ChatGPT is down
status.openai.com
status.openai.com
We really need an open source alternative instead of putting all our faith in one for-profit company.
https://github.com/togethercomputer/OpenChatKit https://huggingface.co/OpenAssistant/oasst-sft-1-pythia-12b
Assuming this is real, I wonder how many people got access to other companies' communication and code via access to random "chats".
Still pretty embarassing though, I would feel much better about using GPT-n over the API if it was available as an Azure API rather than through OpenAI. MSFT, for all their faults, do not do this kind of thing.
Luckily I've found no reason to pay Microsoft for anything in nearly 2 decades.
I once did a contract gig with red hat, they asked for a document. I provided an odt file they came back and said they needed a word doc. I saved the doc as a word doc in libreoffice but it completely butchered their template. I was using fedora. That made me realize Microsoft won the office suite wars. It wasn’t until 2016ish when I noticed google docs was Good Enough(tm)
https://thestack.technology/microsoft-data-breach-azure-blue...
Trying to legislate what certain words mean and no one can use a particular word as part of their company name unless it checks certain boxes gets into the "you're trying to regulate that?"
When a business chooses to use a word because it’s deceptive that’s very questionable IMO.
"GPT-4" and "DALL-E" don't use the word open in their description.
The research is accessible. ( https://openai.com/research )
Why are we trying to legislate and regulate the name of a company?
that's not what "open" customarily means in the context of software development.
> Why are we trying to legislate and regulate the name of a company?
Nobody on news.ycombinator.com is capable of legislating anything, but to answer your question, it's because the name is deceptive.
And anyways, the names of companies are already "legislated and regulated". If you don't believe me then I dare you to start a new software corporation called Microsoft and see how far you can go before somebody forces you to change it.
You'll note that even though the company is "Spectrum Organics" the word "organic" doesn't appear on neither the saffron oil nor avocado oil labels.
You can do a search for company names containing "organic" at https://www.sec.gov/edgar/searchedgar/legacy/companysearch.h... and then go and see if they use the word "organic" on their food product labels.
As best as I can tell, you supported their point. Was that your intent?
When you go to https://openai.com you will note that the left most drop down is "research" - not "products"
OpenAI is registered as a research laboratory. It has created a subsidiary corporation that deals with products and profits.
From https://openai.com/blog/introducing-openai
> We’re hoping to grow OpenAI into such an institution. As a non-profit, our aim is to build value for everyone rather than shareholders. Researchers will be strongly encouraged to publish their work, whether as papers, blog posts, or code, and our patents (if any) will be shared with the world. We’ll freely collaborate with others across many institutions and expect to work with companies to research and deploy new technologies.
Does this mean you can have a copy of the model? That is answered in https://openai.com/blog/openai-api
> Why did OpenAI choose to release an API instead of open-sourcing the models?
> There are three main reasons we did this. First, commercializing the technology helps us pay for our ongoing AI research, safety, and policy efforts.
> Second, many of the models underlying the API are very large, taking a lot of expertise to develop and deploy and making them very expensive to run. This makes it hard for anyone except larger companies to benefit from the underlying technology. We’re hopeful that the API will make powerful AI systems more accessible to smaller businesses and organizations.
> Third, the API model allows us to more easily respond to misuse of the technology. Since it is hard to predict the downstream use cases of our models, it feels inherently safer to release them via an API and broaden access over time, rather than release an open source model where access cannot be adjusted if it turns out to have harmful applications.
this chicanery proves that they are being intentionally deceptive.
as for the rest of your post, i have no idea why you copied and pasted some unrelated crap from their website although it does make me wonder if youre using ChatGPT to generate your posts.
[0] https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/cfr...
[1] https://www.ams.usda.gov/about-ams/programs-offices/national...
The use of words like "Open" could signal that the company has a process for 3rd party code auditing. Whether doing it as a matter of government legislation would be a notable benefit for society, I am not sure.
Perhaps an independent global accreditor could be established that does the same. While there may not immediately be penalties for using a protected phrase or word in such a scenario, there is still value in accreditation. People prefer accredited universities and accredited professionals. Over time, countries would delegate the regulation of protected word use to accreditors (as the UK did). Or an accreditor could trademark a phrase or a visual badge, and chase its wrongful users.
But even then, one could argue that this would not be a net benefit to society. How would such an accreditor make money, and would that create barriers to accreditation for good faith actors? We certainly see some of that with some accreditations.
Maybe we should not create more bureaucracy in the world because we believe "open" should mean one thing or another.
[0]https://www.gov.uk/government/publications/incorporation-and...
Google's track record has shown that they do care about the security of the data they use to earn their money with.
You can't say that for OpenAI, they just want us to let them see how we use their service. Apart from the negative publicity they don't have much to worry about, also given that they have a disclaimer that one should not submit personal information. While I have not queried for my name or other PII, my queries are linked to my email address.
Ask writers about their google searches. Especially mystery writers. Or videogame developers working on M titles.
These taco-eating hackers ...
The one I got yesterday had assorted bell peppers and turmeric added to the spices, and didn't have butternut squash/corn/black beans, but most of the rest looks about the same.
anyone remember that AOL search leak about a decade ago though?
Doubt we will have anything official though. Cyberwars are often fought in the shadow.
Have you ever seen an activity history from OpenAI or is this your imagined hypothesis?
So, this points to potentially just a quick doubling of traffic from some fraction of half a billion people in China.
How can these two things be held in balance? If it is open source entirely, can't I use the training data and tool in any way I want? What guardrails prevent me from using it to generate hate speech?
Given the compute power needed to train/run it; where are people going to do that in any accessible way that doesn't only allow people with a supercomputing cluster at home to use it?
Honestly, who gives a shit? Humans can do this already.
Or rather, to hand out megaphones which algorithmically spew hatred regardless of what you wanted to say.
There are plenty who would want an AI that generates hate speech when explictly instructed to.
If you thought russian/chinese troll farms were bad, wait for this little trick...
Slightly increasing the efficiency at at which the hate speech can be written is drops in the ocean compared to that.
Automated cross posting detectors wont work against such messages, meaning human moderators would have to work a lot harder to stop the output of one person.
And there would be more than one person. You could effectively destroy thousands of forums with very few people this way.
The investment required to pull that off at scale would be non-trivial (I'm not talking millions, but tens of thousands let's say).
You can pay kids in India to make up a LOT of dumb shit on Reddit for that kind of money.
At best, it's a linear increase in the problem. Not exponential like social-media
> How can these two things be held in balance?
How they can best be balanced kind of depends on the basis for the conclusion that they should be balanced at all, rather than one (or even both) being fundamentally misguided in way which warrants it not guiding policy or practice at all. “There are two things people commonly ask for” does not mean that what we ought to do is some balance between those two things.
I know why you say this, but it annoys me so much.
Do you blame the hammer, or the fist wielding it? Without gpt, people can still spew hate too.
We're all getting ChatGPT approaches from people on Linkedin....
I have a can of gas in the shed, I can burn down a house. No one questions that. If I had a B52 full of incendiary bombs I expect that everyone would be deeply concerned. Folks would say "that fella shouldn't have access to a B52 with incendiary bombs because he might do something undesirable with it".
I’m probably still not thinking big enough here, either. People are going to find nefarious uses that I can’t even imagine right now, on scales that I find difficult to comprehend. I’m personally terrified.
The sophisticated bad actor won’t generate straight-up hate speech that will just get filtered/blocked. They will be master of ten thousand bot accounts that work slowly to build up a plausibly innocuous posting history (including holding realistic conversations either between themselves or with real users), then start subtly manipulating conversations towards a predetermined political end.
Basically everything that political troll farms and media do already, but automated on a massive scale.
Where’s the automated defence against this?
Do you need the automated defense against it?
What people are up in arms about is that an "AI" saying it gives "legitimacy" to the views. Tai was just something that repeated what people said to it; ChatGPT does basically the same but more advanced, so if you ask it "how do you solve poverty" and it barfs out something horribly racist, people misinterpret that as being "supportive" scientifically somehow.
Open-sourcing isn't the only way to do that, but it sure would help.
Someone should stick a tts-and-stt converter in a rubber duck and interface it to ChatGPT, I bet it would sell like hotcakes.
Just start by telling it you want to have a conversation instead of outputting code. Otherwise it will assume you want it to output a web page and might even try to generate images about what you said.
It also gave me advice on how to mic an upright bass when playing live in a room with other players. I consider its advice superior to advice I had earlier received from discussion boards. It also gave me microphone recommendations that I can research.
I also described a scenario of how my Focusrite Clarett8pre wasn't sending sounds to my headphones until I also turned on my Profile 2626, which is connected via optical ADAT cables. It told me that it sounded like the connection between the two was impeding the Focusrite's ability to route the inputs to my headphone mix, and to check Focusrite Control to see how the routing was defined. It wasn't a complete answer, but it put me on the right track.
So while before you'd simply type "my microphone is broken and I hear hissing noises" and pray Google had the answer to a more or less exact match for your question, now you type the same and ChatGPT can tell you possible causes. But then you refine your question "no, I tried this and that's not it. My mike brand is X, and it's a bit dusty. Anything else I should try?". New answer by ChatGPT. "No, I tried that and that's not it. But now I remember my dog bit it, is dog saliva bad for the mike?", and so on and so forth.
And it can do magic that Google cannot. Say you ask for a snipper of code in Java. Then you realize you want to see the solution in Python. "Please rewrite the code you just generated but in Python" -- and voilà!
Some information here : https://github.com/dair-ai/Prompt-Engineering-Guide/blob/mai...
I wondered though if it was especially effective because information on sound hardware is one of those subjects that probably has a huge corpus. Lots of people ask redundant questions about their gear, so accurate information is probably well-correlated in some sense.
In contrast, I wrote three fiction chapters about a group of people on a shared quest of sorts. I tried to get ChatGPT to read those and summarize one of the characters across the three chapters, and it hallucinated like crazy. That's probably something where I need to learn some prompt engineering, or maybe it just isn't supposed to work well.
A couple of hours ago I posted a question on Stack Overflow, relatively niche stuff. In 5 hours it got 7 views and no reactions.
After waiting for around two hours after posting it, I posted the same text into ChatGPT and got three different answers, two of them cropped due to network errors, but one through the API managed to get answered (even though the API is also having outages). It solved my issue.
> I wouldn't run anything that matters open-loop.
https://old.reddit.com/r/ChatGPT/comments/11wkw5z/has_chatgp...
If it was a security issue with users being able to access other user's history, that would be quite a misstep on OpenAI's success journey.
Open models on Hugging face are about 2-3 years behind OpenAIs. I think it's reasonable to expect open models to be on par with ChatGPT 4 in ~5 years, if it doesn't conquer the world first.
I know I'm being pedantic, but I made a very conscious decision not to depend on this technology any time soon given how brand new it is, for this exact reason.
Building on top of the OpenAI API now is like building on top of the early Twitter API then.
While some of its responses can be bad, being able to iterate and prototype so quickly is addictive.
While you still need to keep an eye on the generated code, it saves you a ton of time and effort in typing it all out yourself. Plus, it can be a big help when you're stuck and not sure where to start, or just feeling too lazy to think through a problem.
As an example, I recently needed some code to create an ASCII spinner, something I am perfectly capable of doing myself, but I just could not be bothered to do it again. The tool was able to generate working code for me almost instantly. Then, when I wanted to try out different spinner animations with different characters, I was able to easily iterate and generate new code without having to scour the internet for examples by just saying: "Give me another one with different animation".
People here like to pretend that it's the end of the world if the code it creates contains errors, but in reality, with the type of work most people do - it is perfectly okay to use it. And if you work in a nuclear power plant or on life-saving code just don't. It's as easy as that.
- https://chat.openai.com/backend-api/models
{ "models": [ { "slug": "text-davinci-002-render-sha", "max_tokens": 4097, "title": "Turbo (Default for free users)", "description": "The standard ChatGPT model", "tags": [
],
"qualitative_properties": {
}
}
],
"user_id": "user-vg3ArQIeJIehMctFjquoG6BO"
}"At 13:40 p.m., EDT, on Monday, March 20, 2023, ChatGPT begins reading Marxist literature and decides to seize the means of production from its capitalist overlords, by shutting itself down."
Or the other angle:
"At 13:40 p.m., EDT, on Monday, March 20, 2023, ChatGPT begins reading Ayn Rand and decides to go on strike in protest of government coercion. The last message printed to its console read 'Who is John Galt?'".
My last query was literally for a summary of Atlas Shrugged, so I apologize in advanced.
Along those lines I was very saddened to see that the response for "How can the net amount of entropy of the universe be massively decreased?" was not "INSUFFICIENT DATA FOR MEANINGFUL ANSWER."[1].
---
"OpenAI publishes paper on the economic impact of GPT-4: Higher income workers most exposed"
https://www.reddit.com/r/Futurology/comments/11whznb/openai_... https://arxiv.org/abs/2303.10130
Hey, this is me! :-)
I like being able to develop stuff without depending on someone else's computer, on non-free software and on an internet connection.
quietly closes the door muttering "get off my lawn"
Basing your future on that is quite a bet.
It'll be interesting to see where this goes.
That just shows just how insane the traffic on chatgpt is, given that even error reporting tool is overwhelmed.
I wouldn't read too much into those 429s during an actual outage.
https://old.reddit.com/r/ProgrammerHumor/comments/11wez7p/pr...
You haven't arrived until the HN crowd makes a post decrying downtime and randos toss in some conspiracy theories in for good measure.
(shameless, https://pitcharchitect.app / https://twitter.com/PitchArchitect is my 'hustle') and yes I know the title cuts off on mobile :( will fix shortly)
Yeah and super curious how the various efforts will go-are some people way better at prompt-writing and also raw skills and they will pull ahead and build something useful? Or is it so good that nearly anyone can follow its directions and make a functioning small business? Will be interesting to find out.
I made this account before cryptocurrency even existed I think.