throwaway4aday

Why not just write a script that does this but with all of the model providers and requests multiple completions from each? Why have a whole ass editor open just for code review?

I think you need to be more specific about which "chatgpt reasoning model" you used. Even the free version of chatgpt has reasoning/thinking now but there are also o1-mini, o1, o1-pro, o3-mini, o3, and o4-mini and they all have very different capabilities.

They only did that for image generation. The more interesting part is that an LLM can approach or find the correct caption for an image, video or audio during test time with no training using only the score as a guide. It's essentially working blind almost like the game Marco Polo where the scorer is saying "warmer" or "colder" while the LLM is finding its way towards the goal. This is an example of emergent capabilities since there are no examples of this in the training data.

Are you contending that every human derives their reasoning from first principals rather than being taught rules in a natural language?

Doing better than Firefox

It seems that you didn't understand the main point of the exposition. I'll summarize the ops comment a bit further.

Points 1 and 2 only explain how they are able to erroneously justify their absurd beliefs, they don't explain why they hold those beliefs.

Points 3 through 5 are the heart of the matter; egotistical and charismatic (to some types of people) leaders, open minded, freethinking and somewhat weird or marginalized people searching for meaning plus a way for them all to congregate around some shared interests.

TLDR: perfect conditions for one or more cults to form.

That's essentially what R1 Zero is showing:

> Notably, it is the first open research to validate that reasoning capabilities of LLMs can be incentivized purely through RL, without the need for SFT.

Worked great for Vancouver /s

This is a quick "fix" that has a lot of unintended consequences. I've seen it first hand and the only people it benefits are those who are already wealthy. Everyone else gets a lot poorer as their cost of living skyrockets. Homelessness explodes as rents and housing costs increase dramatically, people who were living a humble but decent life before are pushed into poverty, crime both non-violent and violent increases and so does drug use. As far as I can tell the only people that actually benefit from this scheme are landlords and housing developers who slow walk their projects so that they can charge the maximum price per unit. Compared to the previous fairly stable state (which you call stagnation) the locals are much worse off. It also tends to ruin the character of places that were previously seen as a vacation destination for a unique experience, all of that just gets paved over and turned into a bland tourist trap barely different from any other place. Count yourself lucky if you live somewhere that has been passed over by this horrid money making scheme.

I'm not sure if you've read many comments on here but there seem to be more people railing against generative AI than there are those sharing useful and interesting stories or critiques. It's one of the reasons I've been spending less time here, I used to be able to rely on HN for interesting perspectives on both cutting edge and historical tech related topics, often seeing insights from people sharing practical knowledge that was hard to find elsewhere. For the past year or two it's taken a hard turn towards a very spiteful and shallow gripe fest that feels like the same thing that happened to Reddit years ago when it took off in popularity. It doesn't make a lot of sense to me why people are coming here to complain about technology and it just adds a bunch of noise that you need to sift through if you are interested in the tech being discussed. Feels more like people fishing for upvotes by sharing their "hot take" which inevitably a cookie cutter opinion that you see all over social media without any original thought behind it.

Step 1: https://claude.ai

Step 2: Write out your description of the thing you want to the best of your ability but phrase it as "I would like X, could you please help me better define X by asking me a series of clarifying questions and probing areas of uncertainty."

Step 3: Once both Claude and you are satisfied that X is defined, say "Please go ahead and implement X."

Step 4a: If feature Y is incorrect, go to Step 2 and repeat the process for Y

Step 4b: If there is a bug, describe what happened and ask Claude to fix it.

That's the basics of it, should work most of the time.

Not so sure about those examples and pairing with the idea of quick and dirty work.

Your point is on the verge of nullification with the rapid improvement and adoption of autonomous drones don't you think?

I don't think motor skills are a good object to use in an argument about verbal vs non-verbal thinking. We have large regions of our brains primarily dedicated to motor skills and you can't argue that humans are any more talented or capable at controlling our bodies than other animals, we're actually rather poor performers in this area. You're right to say that you aren't conscious of the very highly trained movements you are making because they likely have only a tenuous connection with any part of your brain that we would recognize as possessing consciousness or thought, they are mostly learned reflexes and responses to internal and external stimuli at this point like a professional baseball player who can automatically catch a ball flying at him before he's even aware of it.

Yes! that too.

The tower is also how they plan to perform maintenance and re-staging for another launch. The tower can place it back on the launch structure or lower it down to the ground if it needs to be transported back to an enclosure for extensive repairs but a lot of work is just done with it at the tower. I imagine the goal is to eventually get to an automated system that catches the booster, inspects it for damage, clears it for relaunch, positions it on the launch structure, grabs another Sharship second stage and stacks it and then refuels the whole system and launches as soon as possible.

Plus it's really big https://images-bonnier.imgix.net/files/ill/production/Starsh...

true but don't forget this is test flight #5

not sure how anyone can miss the "throwing away the airplane" part as being the cost driver.

With a payload volume of 8m diameter by 22m height you could fit a James Webb size telescope inside with minimal folding. The sunshield (21.2 m by 14.2m) would only need to fold along one axis and the mirror (6.6 m) could be monolithic instead of having to fold, probably only requiring the mounting points for the primary and secondary to be hinged. This shouldn't be discounted because it makes telescope design much simpler and less expensive.

It also allows for launching individual space station modules that have almost the same volume as the entire ISS in one launch.

Their plans for refuelling on orbit with tanker versions of the starship open up the entire solar system to unmanned missions with much shorter timelines and much higher payload size and weight.

The fact the entire system is re-usable will make it both cheaper and faster to use than any other launch system.

All of this combined mean that it won't just be countries and space programs bidding for space on launches, it puts space within reach of many corporations and some private individuals. This isn't conjecture, it's already happening with the Falcon 9. Starship will make it even more accessible.

You're pretty good at writing clickbait headlines yourself. Those huge checks are a DoD contract for unblockable internet coms not some handout and for $20 million a month it sounds like Elon is giving them a deal, at least compared to what Lockheed Martin, Boeing, General Dynamics, Raytheon, and Northrop Grumman get every year. They don't just suck the teat, they eat half of it.

Unfortunately, they've taken to doing things like saying Horse C is in the lead when it's actually Horse B and telling everyone Horse A has gone lame when it's clearly neck in neck with Horse B while pretending that Horse D doesn't exist because they don't like the jockey.

someone should revive it as a GUI debugger. doing so would narrow the scope which should cut down on complexity. let people use their own editors and IDEs for making code changes. the focus should be on inspecting state, seeing live updates, allowing you to smoothly change inputs and immediately see the output, plus the usual debugger features like stepping through the program.

Wow, great work and kudos for making the source available! I'll have to give this a try.

we're still very far away from complete automation. one person now can do the work of 10 or 100 workmen in the past but people are still needed to transport and set up the feed stock, monitor, configure and service the machines, remove, inspect and assemble the finished products, on and on there are so many other tasks that only a general purpose agent like a human can do. you could construct an assembly line that is so completely integrated that it can almost run untended except for maintenance and recovery from failures but it will only ever produce exactly one model of one product and if you even need to change the weight of one of the parts it produces and then assembles you'll need to manually reconfigure a good deal of it creating significant down time.

it doesn't help that we impede the progress of automation by outsourcing labour overseas where the cost of labour makes manual processes still viable or by importing temporarily cheap labour until they realize they're getting a raw deal and move up the ladder with everyone else.

if we want to develop the technology needed to alleviate the burden of manual labour then we have to disallow these temporary quick fixes. if we do nothing they'll run their course in a few decades anyways, it'll just be a slow walk to the same destination giving the people who are acting in exploitative ways ample time to stuff their coffers with the fruits of their schemes while letting the rest of society rot on the vine mere steps away from the solution that would benefit everyone.

I don't think there's a single way that we learn things, there's too much variety in how, when and why things are committed to memory and still more of a difference with things that actually update our thinking process or world model. We forget the overwhelming majority of sense perceptions immediately and even when we are intentionally trying to learn something we will fail to recall it even a few seconds after we see it. Even when we succeed in short term recall the thing we have "learnt" may be gone the next day or we may only recall it correctly some small number of times out of many attempts. Contrary to that some things are immediately and permanently ingrained in our minds if they are extremely impactful in some way or sometimes for no apparent reason at all. It's too deep of a topic to go into but all this is to say that it isn't so simple as to say that continued pretraining of an LLM is completely dissimilar to how humans learn, in fact the question and answer style of fine tuning that is so widely used to add new knowledge or steer a model to respond in a certain way is extremely similar to how humans learn e.g. quizzing or testing with immediate feedback and repeating the process with many samples that vary their wording while still pertaining to the same information is one of the best ways for people to memorize information.

> it's pretty jarring if anyone can hear it

That's only because you haven't put any effort into sound design. Notification and alert sounds can be horrible like many of the ringtones people choose or they can be pleasant and unobtrusive like certain defaults in various apps. Try browsing some sound effect libraries on various game asset stores, there are many free effects available, and choose something that isn't jarring but is still unique enough to recognize. Avoid sharp beeps and boops or loud melodies, look for ambient sounds like the click of a lock or switch, the swish of paper or fabric, a soft impact sound like dropping a slipper or flip-flop, something you'll notice but won't startle you.

I would think they'd be more likely to form the language centre of a composite AGI brain. If you read through the known functions of the various areas involved in language[0] they seem to map quite well to the capabilities of transformer based LLMs especially the multi-modal ones.

[0] https://en.wikipedia.org/wiki/Language_center

Setting aside land and limiting or not allowing development of it so that it can be publicly available for enjoyment of nature or hunting and fishing is a laudable endeavour.

Banning all human presence and use of large swaths of land, on the other hand, is right up there with medieval Royal forest laws[0] and I suspect it serves the same purpose.

[0] https://en.wikipedia.org/wiki/Royal_forest

I didn't say anything about a timeline for _solving_ these, just that the short timeline for drop off in demand for compute is unfounded since there is still so much ground to cover. The article takes the shortsighted view that the current state of text generation feels like it's in a lull (I strongly disagree with this for a variety of reasons chief among them being that 1. the supposed stall in progress hasn't gone on long enough to call it and 2. the big players are all focusing on productization and making the current SOTA as cheap as possible to improve their bottom line and expand its applicability) but there are a large number of other domains and sub-domains where these techniques can be applied and will likely see similar rapid advances as the amount of available data increases.

While I agree that we don't want to extrapolate too much I disagree that this type of exploration may not benefit from more compute. We won't know until we try and since we have what seems to be a very generalizable architecture it makes sense to take the brute force approach of creating models of that data by scaling the amount of data and the amount of compute we dedicate to it. If it turns out not to work then we've learned something. As it turns out, Logic and Algorithms has seen some early success using Transformers (Searchformer) https://arxiv.org/abs/2402.14083