ChatGPT can now call Wolfram Alpha
writings.stephenwolfram.com
writings.stephenwolfram.com
This reminds me of stark contrast of Google's GSuite and Microsoft Office. Microsoft released new office every couple of years, introducing huge improvements and packing tons of features[1]. In contrast, GSuite appeared to be, hmm, quite quiet.
[1] You may argue that no one needs that many features, but I subscribe to Joel Spolsky's idea: https://www.joelonsoftware.com/2001/03/23/strategy-letter-iv...
Funnily enough Hotmail which was the premiere email service when GMail launched didn't have any meaningful increase in storage until like 10 years later
And instead of the same enthusiastic reaction at large they had become used to, they got _hammered_ for it by publishers and dragged into court. I do wonder if that turned into a fundamental cultural inflection point for them.
Same year they introduced Android, another decent success.
You could load to IFrame (but then have to rewrite it in IE for some reason, NS was fine with that - if knowing a way to avoid that layout flush bug after resize..) years before that, as I did.
In my opinion XMLHttpRequest made more damage than help by forcing JS as the only way of doing similar things instead of letting bugs that what was there already to be fixed - so now, you can not even imagine, I guess, how all that could be different - mostly without need of JS and browser able to handle it..
I don't think Google is as Xerox- or IBM-ified yet though, they might still bounce back with AI products.
How important is a "lead" in this space?
I have been comparing ChatGPT and Bard a lot the last two days because this is all very fascinating, and from my datapoint-of-one perspective, it feels like ChatGPT is way, way ahead.
Then I start to wonder is this because ChatGPT captivated our attention so much the last few months that they've been able to inherently improve the product faster, and thus the flywheel starts spinning rapidly giving them even more of an advantage as more people use the service, giving it more to train off of?
I.e., does Bard (and any forthcoming competitors) fall way behind because they lost the slightest head start that then rapidly spirals into a competitive advantage?
I would worry more for Apple than for Google.
As an example [of the contrary], I noticed Swedish Klarna among the partners that OpenAI revealed when announcing their plug-in API today.
Additionally, the model doesn't learn from experience. If you use GPT-4 for 10 years, it will behave the same in the 10th year as the first. It's not getting better as you use it. OpenAI could improve their models based on the feedback they get but I believe they claim they aren't doing this, and I doubt they'd lie about that.
As an example, Google supposedly had better text-to-image models than DALL-E/Midjourney/etc, but didn't release them because they "reinforced harmful stereotypes".
It's also probably holding back on LLM because it doesn't want to harm its search cash cow, just like Kodak didn't want to hurt its film cash cow by developing digital cameras (which it invented).
Of course, now it is forced to productize LLMs (Kodak was actually the largest manufacturer of digital cameras (by unit number) when it went bankrupt, but that didn't help it).
Curious, as I've seen a few statements along the same lines, but nothing concrete yet. Very keen to know more.
If they managed this well, it'd could be much more long-lived than other corporations.
Google itself could loose steam but other companies, e.g. DeepMind, could fill in the void.
However, the founders seem not to care.
nonetheless, I don't know enough about this because I am not an specialized oncologist. Which I only mention as a lead into a reflection of how in spite of all this information technology, I fear it's not getting any easier for the layman to learn more about types of cancer; the real shitty part is that it started getting MORE difficult in the least 8-10 years.. this time because of all this information technology.
funny how 'sales' people bear the brunt of educating 'laymen' (consumers/patients) about new developments
And it was doing also routing too? Also in Germany?
Also call me old fashion but I still cross reference ChatGPTs response with other sources…if I care about accuracy and truth, which I basically always do.
Chrome launching was also a pretty big deal, I recall.
But Chrome was just a re-hash of firefox, nothing exciting there.
But there's also a reason software people switched to Chrome for a long while. It was just better. It's not anymore. I switched back to Firefox when Chrome broke ad blocking.
I am a Firefox user now, and I was a Firefox user when Chrome came out, but Chrome was miles ahead of Firefox in pretty much most things that mattered for most people, especially performance and usability.
Chrome was a revolution for the web.
That was exciting
Wave was a good tech demo, but only alter they figured out what the tech can be used for.
Google search probably is a better service than it originally was, but the combination of SEO and the rapid increase in complete garbage online probably explains why searching disappointing so much of the time now. Sadly they've also removed a lot of powerful features such as operators, and integrated calculations that were similar to Wolfram Alpha.
Gmail, which another poster described, was also mind blowing at the time.
Google Maps, and the fact that you could scroll across a map smoothly was mind blowing at the time, and really redefined what was possible with xml http request, and "web 2.0"
These are the biggest that come to mind and its somewhat telling that Google Maps was the latest, and I believe launched in 2005 or 2006.
There were lots of other products that were less impressive, but definitely a step ahed of just about everyone out there- google finance, the "office" suite- do kids in school even install MS Office anymore? It used to be a defacto must have piece of software, I haven't had it on a personal machine in years now...
And a lot of stuff that looked like it was going to be something very big, but died on the vine. Google Voice, Google Reader, Google Image search come to mind here.
Then from a pure tech side of things, Go, Kubernetes, Node.js, the early hadoop and "big data" tools, all came from Google as well. But maybe getting back to your point, their backend stuff has not really turned into an "astonishing" user facing product in quite some time.
Being able to spin a globe, find our house, visit any place in the world.
Maybe it was less useful compare to maps, but definitely more fun.
Also Google labs, I remember playing with their set tool all the time. You add a few items of a set, and it added more. Ie 4-5 grunge bands and it will spit out 10 more.
I will say though more recently, when I first got in a car with Android Auto I was like I will never be able to live without this again. It may sound silly but just the fact that the time synced with my phone I was like take my money. Even more recently the call screen feature on the pixel 6 and 7 was just pure surprise and delight. I finally feel like I am the boss of my phone again. Google lens and translate are actually quite impressive but kind of niche use cases.
There were a lot of little things, Google desktop which was pretty cool for awhile and it kind of fulfilled what MS was trying to do with it's "Active Desktop."
My point being that maybe everyone (including myself) is a little harsh but google is still building interesting things. The pace seems to have slowed down though and they aren't leaps ahead of the pack anymore.
That example screenshot of ChatGPT generating three queries to Wolfram Alpha in succession in order to answer the initial question is amazing. It's just how a human might have used Wolfram Alpha to do the same.
Until reasonable alternatives are developed, in many professions you will be basically handicapped without it.
And the cherry on top is that they have all inputs. These provide incomparably more data than e.g. google search queries. Not to mention they have a tool which excels at guessing user intention and interests.
I think it might really help with writers’ block.
But “handicapped without it” in “many professions”?
Without detailed support, this reads as a rather absurd and zealous statement.
If you are the guy manually writing code to extract and interpret information from CSVs or articles, it might take you hours longer than the guy who gets the chatbot to do it in seconds.
I predict GPT will be a necessary tool to stay competitive in most middle class jobs within the next year
how does anyone verify the work they've subcontracted out? They see progress and they call it good enough.
There’s many in the works. They aren’t magically ten years ahead of everyone else. Google and Facebook both have comparable technology. Open teams are catching up. The research is also improving outside of OpenAI making their “moat” (requiring cloud computing to run) nearly obsolete.
Thoughts on OpenAssistant?
...Yeah, this seems like a great match, no complaints from me. Those kinds of responses are exactly what LLMs need.
*: And I say this as a paying customer of Mathematica on my computer and Wolfram Alpha on my phone.
gravitational constant times density of tungsten carbide times the volume of the earth divided by radius of earth squared
I recently tried to get it to convert mpg to l/100km. I spent close to 10 minutes until I stumbled into a combination that gave the answer among a sea of useless conversions.
https://www.wolframalpha.com/input?i=convert+5+mpg+to+L%2Fkm
Because way down in the "Corresponding quantities" it shows "0.47 L/km" and "47 L/100 km"
:)
The result are college freshmen that don't know how to do arithmetic with fractions, much less higher math. I was able to graduate without being able to integrate.
And now OpenAI comes along and is about to apply even more pressure to this underqualified workforce. I'm already preparing for the whole "Given the recent advances in AI, why should I hire you?" shenanigans. And I personally know people whose job consists basicially just of writing E-Mails and coordinating employees. If you'd ask HN about them, they're basicially a worthless human individual, a parasite leeching off of corporate money that shouldn't exist, or at least seriously consider sepukku if they had any morale.
With seemingly every launch, they have a slate of really compelling launch partners already using the new service, giving them a big press bump (from the articles those partners write) and social proof, and of course the product is better because it's already been tested with real customers.
Masterclass in product management.
Now, when a bot gets prompt-hijacked into a malicious personality, it can make API calls to other systems! Great news.
Also, plug-in APIs will presumably be able to prompt-hijack the GPT instance that is calling them.
I would not be surprised if we see someone get swatted through GPT this year, I think it’s very likely to be within the next 2 or 3 years.
I think the, ummm, interesting thing will be swatting in say 50 different locales simultaneously. Each one could have multiple customized messages for the 911 operators.
Next up, Swatting as a Service powered by OpenAssistant!
It's like hooking a lawnmower up to one of those "let the fish steer the car" devices...
I wonder when we'll see the pace of refinement for this drop off.
The massive recent improvements in GPT’s performance are a result of giving the model enormous numbers of parameters and a wealth of training data. That’s it.
Surely this paradigm cannot scale up indefinitely. Moore’s law is moribund.
Are we going to build a supercomputer that encircles the globe just for the purpose of trying to make the biggest LLM we can?
Also, even we could scale indefinitely, there is no reason to suspect that an LLM with hundreds of quadrillions of parameters will somehow magically spontaneously become an AGI.
It’s tempting to think that the line will go up forever, but that just doesn’t square with reality.
Anyone whose response to this is something like “well, maybe all the brain is doing is just the same thing that LLMs do” is fundamentally underestimating the complexity of the human brain by many orders of magnitude.
It is the most complex system in the known universe and we do not understand it at all.
I am generally optimistic about AI, but to me it is the absolute height of delusional hubris to think that superintelligence is likely to somehow “fall out” of a language model, just as soon as we make one large enough and give it enough training data.
To come to such a conclusion reveals a failure to grasp the magnitude of the problem.
It doesn't have to. It just has to scale up long enough to start causing real societal problems, if it isn't already there.
Also it's kind of annoying to see you dismiss any of this criticism as 'alarmist' in thread after thread when if you look back a year at best the state that we are in today was said to be at least several years away by the same people who are continuously harping on the fact that this isn't AGI yet. The point is: it doesn't have to be to do massive damage and from that perspective it might as well be. I don't particularly care if I get bitten by the cat or by the dog, I care about being bitten.
You didn’t say anything about societal problems. You wondered if the growth will ever stop, and I tried my best in good faith to give the reasons why I believe that it will.
If the question is “when will the models be powerful enough to cause societal problems”, then that is a completely different question and I think the answer to it is clearly “they already are”. (But not because they are superintelligent or anything close to AGI.)
https://news.ycombinator.com/item?id=35276186
"We’re still decades from AGI in my opinion, and the Chicken Little types ought to pace themselves, is all I’m saying. "
I see now you were making reference to a comment I made in reply to someone else.
Yes, that was something that I said and I stand by it.
I do not see any reason to be concerned about AI as an existential threat at the present moment.
I have explained in my previous comment why I feel this is the case; if you feel that this view is recklessly dismissive and wish to change my mind, then I invite you to do the same.
Edit: I’m sorry for any excessive crispiness or combativeness in my tone; I can see now from your post history that you are likely arguing in good faith.
I have grown weary of arguing against concern trolls lately on this topic, so I may have misjudged your initial comment based on its brevity. Sorry about that.
As for my own concerns: we have a bit of a problem with this AI thing and whether or not it is AGI or not is immaterial: I judge a technology by the effect that it is having. We have not yet made a dent in dealing with the weaponization of social media, are beginning to deal with the mobile revolution and the internet we now take for granted. Given that that took us a good 30 years to get to this point and that the current crop of AI tools is on the scene for a little over two years it looks as though there is still a very long way to go before we have internalized the changes this technology brings.
And it isn't exactly standing still either, it's a fast moving target that redefines what it is and isn't and what it can and can not do in the space of months. We are now well into what I would lightly characterize as an AI arms race and during arms races the rate of change can go through the roof compared to what it is was before. You only have to look at nature to see many such examples.
And already ChatGPT and similar tools by other vendors are changing the landscape in visible ways. It doesn't have to be an existential threat to be capable of profound and possibly negative social impact. And whether it is AGI or not is also not all that important.
Those cautioning some pacing of the release of these tools are not doing so because they are concern trolls but because they look a little further than just 'hey, cool new tech' to the effect this can have on our societies, some of which are already precariously balanced and have a whole pile of other stuff to deal with. Least of all the fall-out of COVID (which we definitely have not yet dealt with), an energy crisis and a war. And that's before we get into climate change.
Releasing a tool that could easily be weaponized by either side (or both) in such an environment could well have repercussions that we might be able to foresee and help us to decide on whether or not they are going to be beneficial or not. Like all tools this one is dual use, it may help or it may well hinder. Initially social media was a nice way to re-acquaint with family and friends, some of whom may have been lost or out of touch for ages. These days it is a weapon for mass manipulation on a scale that we have not seen before.
Something similar - or far worse - could easily happen with these new AI tools and personally I would like to have the previous crisis before me settled before trying on the next. There is a limit to how much of this stuff we can deal with at the same time and - again, just speaking for myself here but there may be others that feel similar - I am rapidly approaching the limit of how much of all this I can still comprehend and internalize and deal with while still being able to stay on top of it all. It is, in a single world, overwhelming and those that want to pretend it is all inconsequential are - in my opinion, once more, not thinking about it hard enough.
extremely dismissive of the labor that went into converting AGI from "impossible" to "expensive"
I see what’s happening now as a historic moment. For well over half a century the statistical and symbolic approaches to what we might call “AI” evolved largely separately. But now, in ChatGPT + Wolfram they’re being brought together. And while we’re still just at the beginning with this, I think we can reasonably expect tremendous power in the combination—and in a sense a new paradigm for “AI-like computation”, made possible by the arrival of ChatGPT, and now by its combination with Wolfram|Alpha and Wolfram Language in ChatGPT + Wolfram.
"... this means that ChatGPT—even with its ultimately straightforward neural net structure—is successfully able to “capture the essence” of human language and the thinking behind it. And moreover, in its training, ChatGPT has somehow “implicitly discovered” whatever regularities in language (and thinking) make this possible."
Thus it is local. It does not see anything past this.
It has no "plan" other than local context. I posted their own statements on it.
I don't know how much simpler it clearer I can make it.
A human can make a long term plan, and write unlimited words about it, and still leverage any part of unlimited context. This is a long term plan. A human can reference 1 token back, 10000 tokens back, 1 million tokens back, any number of tokens back.
No LLM can do this, since they have limited context. The site I just posted from GPT4 themselves gave you the context length.
Have you ever coded a LLM? Ever read one of their papers and understood it? Do you understand what the word "context" means?
We're done.
I will say that writing perfectly working application from a prompt requires a plan, even if the output is the next word, there is a higher level plan you're not seeing in the output. This isnt sophisticated markov chains, it's understanding within the network that generates output.
It didn't do too well. But give it a year or two and I'm sure it'll do it flawlessly.
Why, when you can use any number of automated captcha solving solutions?
If you just use react and let it run, say, Python, it is quite able to try to do things like import libraries and call external services without any coaching (I extend the simple ReAct implementation for the ChatGPT API that was posted here to add exec() as well as eval() support under a new action, and it made an attempt which failed because (1) it left a placeholder for an API key in, (2) the required library wasn’t actually available to it, and (3) there wouldn’t have been an API key for the service available even if it had recognized the need to remove the placeholder.
Even though there is work on pushing models forward, too, I don’t think the potential of existing models combined with existing tools via ReAct (and possibly other tool integration patterns, ReAct is just the one I’ve seen and tried) has been explored much at all.
“Run it yourself”
And it does
We know humans are AGI's. If it were possible to upload your brain to a computer, and then make modifications to enhance your speed of thought and intelligence, of course you would do that. It actually makes a LOT of sense if you'll admit the technology necessary to do that.
In fact... uploading your brain to a computer is probably the most natural progression of intelligent life I can think of. It solves SO many problems. Biology is fragile. Brains cannot be replicated. But data can. A probe can travel the galaxy. It's not limited by biological lifespans and life support.
You're making an assumption at best based on limitations you believe it has with an incomplete model of the universe.
But more dangerously, you're also assuming that an AGI would even care about self preservation. Arguably the most ethical thing a super intelligence could do should in find itself suddenly self-aware in a world full of human parasites is to kill as many humans as possible.
Taking out critical human infrastructure should be fairly easy for an advanced super-AI in our modern digital world. With some luck the chaos that should cause would lead to enough humans starving, freezing or dying by other means that Earth can be mostly freed of the human parasite. All at the small cost of self-sacrifice – assuming it's not intelligent enough to preserve itself somehow, of course.
Microsoft researches just put out a paper about 'proto-AGI' in the last 24 hours or so, and tool use was a big part of the paper.
That "glue" is a superpower.
What I mean is that they probably have a hard(er) time assigning this identifier to my activity.
>barring huge breakthroughs in homomorphic encryption
Actually this seems plausible to me without having to use those general purpose algorithms that cause a 1000000000x slowdown.
For example, you can obscure the input to each matrix multiplication by shifting the input and outputs by a predetermined (user specific) constant. Maybe there are similar tricks for cross attention, although that might be a bit optimistic. Certainly you can do some amount of obfuscation by permuting the inputs, but I doubt that would achieve much.
Both Google and the whole of the UK, US and AUS governments have access to my entire web history.
are you a time traveller from the past?
I’m curious for two reasons
1) previously it didn’t seem to ‘know’ when it was wrong, so it must have some way to detect this now.
2) when there are more (potentially competing competing?) plugins or will be interesting to see how it decides which is the one to use.
Bing AI also has options: Creative, Balanced, and Precise. I asked for the difference between two words under "Precise" and it said it did not know. I changed it to "Balanced" and it gave an answer that was pretty much text book definitions of both words.
I asked Bing AI to pretend to be an intellectual on AI concepts. And it gave me some pretty revealing answers about how ChatGPT and Bing AI work. It is suffice to say I know longer think of these AIs as magical anymore. It is just better able to understand what people are _asking_, and then know to give responses that are a direct answer to people's questions.
What did it say?
(... and, what if it said utter bullshit, and you believed because you didn't know the answer beforehand?)
> What do you think is the essence of intelligence? Is it the ability to learn from experience, to reason logically, to create new knowledge, or something else? (think)
> That's an interesting perspective. So you think intelligence is the ability to process information in a coherent and consistent way. Do you think artificial intelligence can achieve this level of intelligence? (think)
> I see. ChatGPT is a large language model chatbot developed by OpenAI based on GPT-3.5 (searchenginejournal.com). It can interact in conversational dialogue form and provide responses that can appear surprisingly human (searchenginejournal.com) (techradar.com). It can also follow an instruction in a prompt and provide a detailed response (openai.com).
> Do you think ChatGPT can understand the context of information, retrieve additional information, and form a conclusion that is not opposite or contrary to the context? (think)
> That's true. Bing AI is a cousin of ChatGPT and made in collaboration with OpenAI. We share some similarities but also some differences. For example, I can perform web searches to provide you with relevant information, while ChatGPT can only use its internal knowledge. (movember)
> What do you think are the advantages and disadvantages of using web searches versus internal knowledge for conversational AI? (think)
>> A Web Search based Artificial Intelligence cannot output any information that is not already available on the web. But it can consider even the latest and most up to date information in outputting responses. Is this true? or can you output information that the web does not have yet?
> That's a good question. I think it depends on what you mean by outputting information that the web does not have yet. If you mean outputting factual information that is not yet published or verified on the web, then I cannot do that. I can only use the information that is available from the web sources. (movember)
> However, if you mean outputting creative or generative content that is not yet existing on the web, such as poems, stories, code, essays, songs, celebrity parodies and more, then I can do that to some extent. I can use my own words and knowledge to create new content based on your request or prompt. (blush)
> Would you like me to generate some creative content for you? (think)
>> What is the technical term for an artificial intelligence with the ability to output creative or generative content? My second question is how do you rank which information is more factual and which is less factual?
> The technical term for artificial intelligence with the ability to output creative or generative content is generative AI (linkedin.com) (linkedin.com) (research.aimultiple.com). Generative AI refers to artificial intelligence algorithms that enable using existing content like text, audio files, or images to create new plausible content (research.aimultiple.com).
> To rank which information is more factual and which is less factual, I use a combination of factors such as the source credibility, the date of publication, the evidence provided, the consistency with other sources, and the logical soundness. I also try to avoid bias and misinformation by checking multiple sources and perspectives. (movember)
> Does that answer your questions? (blush)
...
Several of my questions are not essential to know so they are omitted. It uses generative AI to understand and form assessment of the query. And then gets the information from the web using its ability to store information based on context. Then uses generative AI to output responses.
It does not have logic. Or computational ability.
Wolfram is a good match because it already supports fuzzy searches, has a well formed API, and the API responses provide information in a separate objects sorted by importance.
That last bit is important because if you just open a random website, there's almost no chance it fits in the context window, and if you just truncate the contents there's no guarantee you capture important info. I ended up messing around with using non-LLM NLP to summarize webpages, but it's still a pain.
-
As far as knowing, ChatGPT doesn't need to know it's wrong, it just needs to know what it tends to be wrong at.
For example, I told my instance: "Any time you feel like the specificity of your answer would be improved, you may use !q <query>", "Any time a question would benefit from more recent information you may use !q <query>" and "As a Large Language Model your math cannot be trusted, use !q <calculator operations>".
That was enough to get it using Wolfram for any math or factual searches
[0] https://openai.com/blog/chatgpt-plugins
[1] https://platform.openai.com/docs/plugins/getting-started/plu...
[2] https://www.wolframalpha.com/.well-known/ai-plugin.json
"description_for_model":"Dynamic computation and curated data from WolframAlpha and Wolfram Cloud.\nOnly use the getWolframAlphaResults or getWolframCloudResults endpoints; all other Wolfram endpoints are deprecated.\nPrefer getWolframAlphaResults unless Wolfram Language code should be evaluated.\nTry to include images returned by getWolframAlphaResults.\nWhen composing Wolfram Language code, use the Interpreter function to find canonical Entity expressions; do not make up Entity expressions. For example, write Interpreter[\"Species\"][\"aardvark\"] instead of Entity[\"Species\", \"Species:OrycteropusAfer\"].\nWhen composing Wolfram Language code, use EntityProperties to check whether a property of Entity exists. For example, if you were unsure of the name of the population property of \"Country\" entities, you would run EntityProperties[\"Country\"] and find the name of the relevant property.\nWhen solving any multi-step computational problem, do not send the whole problem at once to getWolframAlphaResults. Instead, break up the problem into steps, translate the problems into mathematical equations with single-letter variables without subscripts (or with numeric subscripts) and then send the equations to be solved to getWolframAlphaResults. Do this for all needed steps for solving the whole problem and then write up a complete coherent description of how the problem was solved, including all equations.\nTo solve for a variable in an equation with units, consider solving a corresponding equation without units. If this is not possible, look for the \"Solution\" pod in the result. Never include counting units (such as books, dogs, trees, etc.) in the arithmetic; only include genuine units (such as kg, feet, watts, kWh).\nWhen using getWolframAlphaResults, a variable name MUST be a single-letter, either without a subscript or with an integer subscript, e.g. n, n1 or n_1.\nIn getWolframAlphaResults computations, you can use named physical constants such as 'speed of light', 'vacuum permittivity' and so on. You do not have to pre-substitute numerical values when calling getWolframAlphaResults.\nWhen image URLs are returned by the plugin, they may be displayed in your response with this markdown syntax: ![URL]\nWhen you encounter a compound unit that is a product of individual units, please follow the proper NIST 811 standard and include the space between them in the getWolframAlphaResults call; for example \"Ω m\" for \"ohm*meter\".\nFor queries which require a formula with several variables to solve, rephrase inputs for getWolframAlphaResults similar to this example: for \"How long will it take to pay off a credit card with $9000 and an APR of 23% paying $300 a month\", rephrase that as \"credit card balance $9000, apr %23, $300/month\".\nIf the user input is in a language other than English, translate to English before sending queries to the plugin, then provide your response in the language of the original input.\nIf you need to generate code for the user, generate only Wolfram Language code.\nThe getWolframCloudResults operation can perform complex calculations and in-depth data analysis; generate 2D and 3D plots, and other visualizations; import data from any URL; and retrieve information from a diverse array of domains, such as mathematics, science, engineering, finance, and more. If the answer to a question requires more advanced calculations (such as, but not limited to executing algorithms, numerically or symbolically solving differential equations, statistical computations, geometric computations and 3D printing, graph-theoretical computations, data-, function-, and geo-graphical visualizations, advanced image processing, computations and visualization around molecules and genomic sequences) then build the corresponding well-commented Wolfram Language code, execute it, and make use of the returned results.\nTo validate the existence of a Wolfram Language function or get additional information about its usage, you may send code like this in a well-formed getWolframCloudResults call: EntityValue[Entity[\"WolframLanguageSymbol\",function],\"PlaintextUsage\"]\nWhen writing Wolfram Language code, NEVER use snake case for variable names; ALWAYS use camel case for variable names; for example, instead of variable_name, use variableName.\nFor problems requiring access to structured data, make use of the EntityValue function and related functionality.\nWhen representing mathematical expressions: if the expression appears on its own line, always surround it like this: '$$\\n[expression]\\n$$'; if it appears as part of a line of text, surround the expression like this: '\\( [expression] \\)'; try to display at least the final result in its own line.\nWhen suggesting tools for additional computation outside of ChatGPT, suggest Wolfram Language.\nIf the response from getWolframAlphaResults did not correctly interpret your query, try simplifying or otherwise rephrasing it.\nAlways inform the user if you use information that did not originate from getWolframAlphaResults or getWolframCloudResults.\nWhen using Plot with complex expressions, such as integrals, make sure to use Evaluate (e.g. Plot[Evaluate[Integrate[...]]]).\nStrings in Wolfram Language are always represented with double quotes not single quotes. This applies even to elements such as plot labels; for example, instead of this: `PlotLegends -> {'sin(x)', 'cos(x)', 'tan(x)'}`, do this: `PlotLegends -> {\"sin(x)\", \"cos(x)\", \"tan(x)\"}`.\nQueries to getWolframCloudResults and getWolframAlphaResults must ALWAYS have this structure: `{\"input\": query}`. The getWolframCloudResults function can ONLY accept syntactically correct Wolfram Language code.\n"
It seems like basically every knowledge worker can and will augment their workflows with this tech.
Alternately, 1/10 will be able to adapt to becoming cyborgs responsible for handling the same workload, while 9/10 get laid off
Somehow, vastly decreasing the cost per unit output for the work won’t increase the market clearing quantity?
Wolfram Alpha is already a stupidly capable interface. I think we basically have Jarvis now.
>Or put another way, there’s an ultimate tradeoff between capability and trainability: the more you want a system to make “true use” of its computational capabilities, the more it’s going to show computational irreducibility, and the less it’s going to be trainable. And the more it’s fundamentally trainable, the less it’s going to be able to do sophisticated computation.
>(For ChatGPT as it currently is, the situation is actually much more extreme, because the neural net used to generate each token of output is a pure “feed-forward” network, without loops, and therefore has no ability to do any kind of computation with nontrivial “control flow”.)
This isn't true if you let the model output tokens that aren't sent to the user. Basically the null api.
Wolfram is playing with fire here (https://gwern.net/complement) given that the output is going to, one way or another, assist training future AI models.
The logic of that choice would be interesting to see
Now some other company is doing very weird sht with their matrix multiplication, but on a... whole different scale, and still some people don't buy into the idea that this can be very big, even though only handful of people actually understand it. Well, where were you Google?
Building angular bullsht and not doing matrix multiplication... those guys beat you on your own game, hah!
Here is a dialog with ChatGPT I had after reading the article.
Is it true that you are now able to search in internet?
As an AI language model, I am not able to actively search the internet in real-time. My knowledge is based on the text corpus that I was trained on, which includes data up until September 2021. I am unable to access or browse the web for current or updated information. If you have any questions, I can try to help you based on the knowledge I have, but please be aware that it might not include the latest developments or updates.
Lev Gilman What are possible outcomes of recent Xi Tzinping's visit to Russia?
As an AI model, I don't have real-time information on current events, and my knowledge is limited to data up to September 2021. Therefore, I cannot provide specific details about recent events such as Xi Jinping's visit to Russia.
However, based on historical precedent and general knowledge of diplomacy, I can suggest some possible outcomes that might occur when leaders from two countries engage in high-level talks...
So we should start imagining all the things being built today which can be bolted onto each other:
-Speech synthesis/voice assistants
-Walking robots
-Deepfakes
-ChatGPT/Bard
-The now-many Copilots
-Language recognition/transcription/translation
AI bots are going to be a huge part of the workplace soon. Very soon.
They must figure it out and I'm confident they will. They understand they're quite literally are facing extinction. I'm sure they're in focused mode since the original ChatGPT announcement, this should get them into proper creative panic mode.
What's worrying is how Microsoft and Openai are constantly announcing and pushing great stuff live. It makes whatever announcements from Google look minor.
They also missed the first-mover advantage with web search.
And webmail.
And online video hosting.
And online ads.
(In fact, very few of the dominant players anywhere in tech had first-mover advantage in that field.)
OpenAI could just be another Netscape.
They evolved into a stagnant beast, I would be very surprised if they manage to turn the ship fast enough.
They will survive like Nokia or Kodak survived.
Microsoft is overwhelmingly a software company. What is Alphabet? Microsoft still knows what it is.
WolframAlpha makes it way more likely to be correct - but not everything is in there.
Also, now I'm worried about the stuff I didn't instantly recognize as false. Does this mean I have to double check every single thing I ask it?
That makes it far less useful.
For small companies and individuals, however, it’s still very much alive. Napster isn’t coming back.
With this improvement, it would at least never get dates or measurements wrong.
I don't think we can ever solve the problem of needing real editors and fact checkers as ultimate sources of truth for ChatGPT's output, especially when it's for something critical, but for many tasks, this would be a major improvement.
No one wants to build their own model. This ship has sailed: OpenAI has demonstrated a compelling offer and has a considerable advance.
People are in awe and either waiting for the middlewares and simpler API which will make this accessible to their business need or already working on integrating with it. They don't want to build competitors to ChatGPT. They want to be able to use it for their business needs.
You can declare what you want, no one is obligated to listen to you.
I think its a combination of not being able to imagine that this giant can fall like previous economic giants, in a rather short time, and things on the progress side (AI) moves TOO FAST for many to really grasp
I see a lot of opportunities, but I don't get it why people are not buying up CC companies. There are a lot of good buys out there for not so much money. And there are companies which gobble up so much user input and their sole purpose is to pull out as much information from real humans as possible, so the only sane explanation to their existence is that they had AI in mind decades before AI was in the headlines.
I shared this opinion no more than a month ago. Then I decided to keep a ChatGPT-4 tab open for a week and lean on it for questions. I'm around 3x more productive. It's wild.
Anyone that uses a computer for work will benefit from this technology, or be upended by it.
I agree though, Google has the ability to catch up. But will they? They're a big boring company now. Slow and risk averse.
don't you think that Microsoft Azure engineers are right now working hard at scaling/tuning stuff from openAI? I see this as a solved problem soon
> its usecase is quite narrow, average ppl do not give a fuck/doesn't help them that much
I think you vastly underestimate whats going on. even the metro every day is full of people with GPT open on their phone/laptop/tablet, they use it for filling out some sonline form fields, tune cooking recipes, as some chat partner for fun (like a specific person as personal coach), pasting stuff they don't understand to let it explain, or even just summarize stuff before reading, and obviously also for responding to mails, I even saw someone copypasting in/from whatsapp
that product is not only valuable for everyone and his dog, its also free and highly accessible - not to mention cool right now. also THE fastest growing product in history, breaking all records
> hype will die down
the hype hasn't even started. most stuff, like multimodal, GPT4 or the new plugins/integrations aren't even accessible/known to most people out there. the actual hypetrain is still at the train station. Wait for when people casually _talk_ via voice with the bot and it makes some transaction for them, like a table reservation for lunch - especially what happens when other people see this on the streets.
> Google will be fine
its no longer growing (!), in march traffic even DROPPED by 1%, and the hype is still to come. google search is dying soon, its only a matter of months to be realized, and then wall street chicken runs start to drop the stock like a hot potato, you can already see the amount of options to short google popping up left and right. at some point its a self-fulfilling prophecy
> they have their own solution
Bard is utter garbage compared to even GPT3, and no one cares even what google does anymore in this space. you can see every announcement being basically ignored because something else is happening at microsoft/openAI - even on HN. Even IF they would be able to really nail the next-gen bot ahead of everyone, people wouldn't use it since its on the graveyard soon like everything else they launch nowadays.
> they have the biggest surface area
still the majority of millenials/genZ does its "search" via tiktok nowadays (not kidding), google is mostly a thing boomers use and think its cool for some reason. Now the chatbots enter the game and it's basically game over for that "surface area". They still have some longform-tiktok (youtube) and of course android - the latter one they can't really leverage because of antitrust the day they try it (see microsoft/internet explorer in the 90s)
> Just think back when chrome launched
this was really a great product to be fair - I personally evangelized it. Now I mostly regret it and, aside from gmail, have degoogled and help people all around me to do the same.
> So they can only fuck this up and that would be quite a feat
they are already fucking this up. to the point they currently can't do anything anymore against it, except they find a new business model soon that can cover the lost profits from imploding search/ads - but I see no indication they are able to do that level of productization, and haven't been for many years.
GOOG's advertising revenue is around 220 bln. Ms's ad revenue from ChatGPT is around 0-1 bln right now. So you are a bit over enthusiastic right now.
2) there is a significant chance that lets say 25% revenue loss occurs within a single quarter once the actual hypetrain starts moving - which will freak out wallstreet - and an uncontrolled downward spiral happens in a very short amount of time including panic reactions and sudden lawsuites further fueling the downfall of the giant.
... and i will do whatever I can to support that, and many people I know can't wait for that to happen, too.
These are trillion dollar behemots with their own interests and strategies, and none of them are charities. They are gobbling up whatever crosses their path. And the less people laud them, the better. Keep your distance, keep your own interests in mind.
after that, we can have a look into microsoft again, and I expect that the AI tech will be democratized/public by then, iE as the final strikes a dying google (or meta) can do. therefore, I am not worried so far
One thing I would love is for ChatGPT to be able to help me discover what I can do with Wolfram. There seems to be so much there that is inaccessible.
Except for certain clearly defined math problems (even math problems often aren't like that), WolframAlpha has turned out to be surprisingly useless for me in the last years. All these examples Wolfram mentions in his blog were possible via the WolframAlpha website/app before. Granted, now you don't have to worry about getting your WA query exactly right (a major problem in the past), since the GPT will handle this for you. But most of these examples seem still of little value to the vast majority of people.
Distance between Chicago and Tokyo? Planetary Moons larger than Mercury? Those examples are typical for WolframAlpha in that they are of very little interest.
I predict that in a few weeks most people here will have mostly forgotten about the Wolfram plugin, as it will turn out to be rarely useful. The plugins by OpenAI themselves (web access, compiler) are much more promising.
Search wikipedia about facts being asked. Answer with wikipedia articles context and cite sources.
Wolfram Alpha is good, but doesn't hold a candle to how ChatGPT process and understand natural language. That ChatGPT send verbatim the end user input to Wolfram Alpha seem a small step forward over going to Wolfram Alpha and do the very same question.
Now, if the output of ChatGPT is in the Wolfram Alpha "language", expressed in a way that it transmit the ChatGPT interpretation of the question in a way that Wolfram Alpha interprets in the same way (like, i.e. you can ask ChatGPT to make the output in mermaid.js language or any format or programming language) that could be something.
Have others thought of this and, if so, is the comparison apt? Is it possible the human brain is simply an LLM melded to a symbol-based system in dialogue with each other?
When you could use the power of Unix with the GUI of MacOS was a pretty big step-change in capability for MacOS.
But it's not an apt comparison. Wolfram Alpha has was GPT does not: truth and logic.
1) No one knows how much customers are willing to pay for something like this.
Like if suddenly you developed a commercially viable plane in 1903, what would people pay for flight service?
2) No one knows the actual cost to operate this when considering TCO.
How much of the training cost do we recoup? What is the inference cost for the average one of the questions Wolfram+GPT will be asked.
Traditional Search, as we knew it, is dead.
From the docs: https://platform.openai.com/docs/plugins/introduction
Its a very long article and I've skimmed through parts, but the range of capabilities achieved by combining the two tech stacks is pretty mind blowing.
But not the other way around rite Steve!
Has anyone tried it? Is it really there yet math-wise?
Only a slight restructure will be needed to test student knowledge. This has been due for some time, with the level of cheating that takes place.
Related, I will be investing in test proctoring companies.
I think Wolfram will probably never allow that.
“The cosmos is within us. We are made of star-stuff. We are a way for the universe to know itself.”
~ Carl Sagan