motbus3

Have you heard about Argentinean escraches?

Any french would care to explain?

Tell me in 1 year about it.

If this is real, this is ridiculous. This would make most internet businesses even more unaccessible

It is despicable that large language models have depleted the content of Stack Overflow. Even before the advent of LLMs, online communities were plagued by individuals who would pointlessly complain and downvote questions or answers, as well as those who would post low-quality content solely to farm points, sometimes even using sock-puppet accounts to inflate their own metrics. It was great when they offered prizes and similar incentives, but they were essentially seeking free labor; once they established themselves as a primary reference, they discontinued the practice, which is understandable.

The only thing community driven I used nowadays is for cooking because LLM recipes are always the most generic thing. (Though I use LLMs for converting metrics or question if something goes wrong)

I feel it is a bit naive to think Jansen would be a spy. He would hire all the spies and be the boss of both sides. It is not like he has to choose it

*what they did to coding

Software engineer is far from be solved

You can't say they are spending 10s of millions on that. First, you cannot calculate their usage cost by what they charge you. Second, if the prompts from the math researches were not there they would not be able to solve the problem, and they know it, and that's why they are trying to silence people or make noise enough so people don't realise what they are doing

Your comment about academics is not based. If you take the time to read any of the papers they published you will see most if not all are pure garbage nonsense.

Read Terence Tao about ai, or anything about the "supposedly" improvement in the FFT. Their claims are all nothing burgers.

I think OP is saying that the work should not be credit to OpenAI. They are basically rehashing the research from someone else. Previously with the fluid problem they sang glories to themselves only to be exposed. They not only tried to shut up the researches but also displayed several concerning behaviours.

There is nothing that guarantees now that they will not steal your ideas and publish themselves.

By now I want them to be sure on what they say they did before they claim they did. It all sounds like propaganda to me. From the I dont know how many works they published earlier, almost none was a significant improvement or proof. And the one that was, was clearly stolen from someone.

I dont know why it is taking so long for investors to catch up with this suspicious behaviour

I think you are on to something maybe. I also think there is a key part where Microsoft and others would hire people so nobody would. I know cases where people were hired and stayed on irrelevant or useless projects for years. Because of non compete agreements this means a good professional would stay few years out of the hands of the competition.

I think now LLMs might play a part of those dismissals but also I think those companies are embracing for thin cows season.

The record profits and etc are the last few for a while I am guessing. My guess they might have up to a year of positive numbers as consequence of deals and layoffs up to this moment.

I think at some point it will be clearer what are the real productivity gains of LLMs. My opinion is that there is a gain but we are already on the phase where there is a deceleration on the improvements and the costs have not yet caught up. The reason for this is the frequent change in the reporting of results. Now they hide the models, they hide the harnesses, they hide the results. They would not be doing that if they were 100% confident it was working perfectly. Also they have been putting too much effort on blocking other initiatives by hiring important researchers with no clear proposition on what they will be doing. One might argue that when you hire Karpathy you never know what he will might be doing anyway, but I digress. We also see some of the huge investment funds starting to switch strategies.

Unless there is no breakthrough, we might see LLMs results in a couple of years. The company I work for and others from friends I know have all started restrictions on costs. Some even started hiring people back.

My prediction is that the low AI cost per token era will end very soon when investors realise that the market is broken and they want their money back for leverage themselves or other businesses.

I even think this is why Elon Musk is merging all his companies. He might making himself artificially richer, but I think he is focusing his efforts to smart weapons and robots.

I think there will no way back to the inflation of people as in 2020/2021 but probably we are going back to early 2010s number.

Not saying that this alleviates anything, but Grok was doing even worse up to the other day. We need to remember to use the same rule for both cases

Agree. I think slowly companies and governments are ditching US tech. Many companies relied on US tech rather than Chinese with worried of CCP intervention and now it is clear that it will be the same with US tech.

I understand that US is migrating to a no friends policy, but I don't think its internal market is nearly ready to support it if clients start to go away.

Anecdotally, I talked to people in a recent event and the common topic was "what we are doing to avoid a blackout of US tech." People on big companies have reserved hardware and alternative software and are scaling down Amazon and Microsoft solutions at a fast pace. People on smaller companies are doing similar strategies but relying on a multivendor strategy.

I think this was inconceivable 3 years ago. Every company in Europe and Asian would gladly pay for amazon for cloud services and now they are using AI to migrate out the lock-ins.

We are seeing American LLM companies more and more relying on "secretive" methods to increase scores and this can only mean that they are not evolving fast enough. The gaps between the reported benchmarks and measured ones are increasing. It might be "safety" concerns or whatever, but I don't buy it.

Like Musk, Altman and Amodei have been overpromising every time they need more money and each time they need more money and now they internet is poisoned with their own content and they can't even trust any sources anymore.

I think if you do iterate on cycles of development and hardening maybe this can be done, but LLM code, specially by Claude, has been too verbose for me. Other models are less