rafaelero

I feel like you are trapped in the first assessment of this problem. Yes, we are not there yet, but have you thought about the rate of improvement? Is that rate of improvement reliable? Fast? That's what matters, not where we are today.

god they keep raising prices

Let's definitely wait for replication, but I am honestly not that surprised that it works. I am surprised it took so long for people to give it a real try. It's such an ideal scenario: every experiment is conducted inside the computer, so there is no need to gather data in the real world, which is the pain point for most experiments in science. The LLM is therefore free to try a lot of different combinations and learn in real time what works and what doesn't.

If they do the same for the coding model they will have a killer product.

Both sides are not equally wrong, clearly. Until yesterday prediction markets were saying the probability of an AI getting a gold medal in IMO in 2025 was <20%. So clearly we should be more hyped, not less.

Indeed it's a very unsophisticated and obnoxious audience around here. They are so conservative and unadventurous that anything possibly transformative is labeled as hype. I feel bad for them.

Gary Marcus is so systematically and overconfidently wrong that I wonder why we keep talking about this clown.

> If it’s mostly a social phenomenon in your company that non believers are pushed out, that does not seem like a fun place to work.

It's not really intentional. It's just that he's slower than the rest of the team and it is noticeable.

I have been working in software for about 6 years and we are working in a new startup building our product. LLM's have been fundamental and it makes things so much faster.

Right now, I’m noticing how my colleagues who aren’t very comfortable using LLMs for most of their work are getting sidelined. It's a bit sad seeing them struggle by not keeping pace with everyone else who is using it for ~90% of our tasks. They seem to really care about writing code themselves, but, if they don't pivot, things are probably not going to end well for them.

So is LLM inevitable? Pretty much if you want to remain competitive.

This should be used for high entropy tokens during pre-training.

2025 and we are still discussing if LLM's are intelligent or not, gee.

Maybe it's your understanding of reality that it's flawed.

So why did you feel the need to post that next-token prediction is not the reason this behavior emerge?

It requires only a little bit of imagination to figure which industries are most likely to be disrupted by AI. If you calculate the size of those industries, the upside is huge. Let's pick transportation and healthcare: we already have Waymo and probably in the future Tesla offering competitive autonomous drivers, and our current LLMs already surpass doctors in diagnosis accuracy. So why, I ask, why would you possibly think those industries won't be disrupted? Do you really think people will stick with mediocre doctors if they can use AI? Or that they will pay a premium for a human driver? Come on.

They should sell for a price it would make pirating it pointless. Like what Spotify or Netflix did to audio and visual content. Then they can use the exposure to find other ways to make money.

Intelligence is also not a skill, but the thing that makes you skillful in all cognitive tasks. Just like what height does to basketball players.

A [meta-analysis of all RCTs](https://pubmed.ncbi.nlm.nih.gov/28526025/) seems to suggest there's no real difference between these kinds of fats. Is there a particular reason you are using only one study to draw strong conclusions?

Any RCT's showing it is incredibly more dangerous than butter or other vegetable oils?

They do work, mainly for keeping a remissed cancer at bay.

But if disabled workers are less productive shouldn't we expect that?

They should incentivize homes to install batteries. That could absorve energy when price is low and relieve pressure when price is high.

But they say they will never produce a better model than what is available in the market.

It's probably not possible, which makes all these initiatives painfully naive.

Oh god, one more Anthropic that thinks it's noble not pushing the frontier.

It's what surveys are showing. Why would they lie?

Do you have access to ChatGPT? You can try it yourself (you can query GPT-4o and GPT-3.5 afterwards). The difference is large enough to be noticeable at least 80% of the time in my experience.

> So, yeah, GPT4 is much better than GPT3.5, but it's got a long way to go before it becomes real AI.

I don't disagree. Just wanted to highlight the poor intuition of the guy doing the video.

More than 60% of programmers are already using it and finding it useful. Boston Consulting Group conducted a trial and concluded that it increases quality of their consultants work while decreasing the time it took to finish it. So no, you are not correct that hallucination makes it impossible to use.

Even more people seem to struggle with the concept that it may be underhyped.