I understand he wants to deflect liability from his platform, but I guess I have to concede that it seems like a legitimate defense. We allow the stock market to exist even though insider trading can happen and it's (I think?) not Nasdaq's or NYSE's responsibility to pursue that. We have a legal system for that.
I think there is still the debate to be had whether prediction market enable too much criminal activity and insider trading compared to traditional stock markets and therefore need to be limited for pragmatic reasons (i.e. the legal system can't keep up), but that's a different discussion.
I don't know. Practically, LLMs are already better conversation partners on any topic compared to the average human I have access to. This also holds in reverse, of course - if someone wants me to explain something, usually they'd be better off asking an LLM.
I think in domains like Math and Software Engineering, they are less constrained by training data anyway. They can synthetically generate and validate programs. To what extent that scales into novel insights is a different matter, but I think they dream of the AlphaGo Zero moment at least in verifiable domains.
How can it ever play against itself on novel software tasks? First it has to come up with the task. Then it can write tests but then it needs to verify that the tests are correct, mixture of experts can come to wrong conclusions, etc...
In addition to the managed interface for agent configuration and so on, is the novelty that all the agents run on Anthropic's infra? Sort of like Claude Code on the Web? If so, interesting that they move up the stack, from just a provider of an intelligence API to more complex deployed products.
I don't want to be overly cynical and am in general in favor of the contrarian attitude of simply taking people at their word, but I wonder if their current struggles with compute resources make it easier for them to choose to not deploy Mythos widely. I can imagine their safety argument is real, but regardless, they might not have the resources to profitably deploy it. (Though on the other hand, you could argue that they could always simply charge more.)
I would have not believed your argument 3 months ago but I strongly suspect Anthropic actively engages in model quality throttling due to their compute constraints. Their recent deal for multi GWs worth of data center might help them correct their approach.
It's very interesting to me how widespread this conception is. Maybe it's as simple as LLM productivity degrading over time within a project, as slop compounds.
Or more recently since they added a 1m context window, maybe people are more reckless with context usage
It has nothing to do with the context window. Reasoning brought measured approaches grounded with actual tool calls. All of that short-circuits into a quick fix approach that is unlike Opus-4.5 or 4.6. Sonnet-4.5 used to do that. My context window is always < 200K.
>Models are not "degrading". They're not being "secretly quantized". And no one is swapping out your 1.2T frontier behemoth for a cheap 120B toy and hoping you wouldn't notice!
>It's just that humans are completely full of shit, and can't be trusted to measure LLM performance objectively!
>Every time you use an LLM, you learn its capability profile better. You start using it more aggressively at what it's "good" at, until you find the limits and expose the flaws. You start paying attention to the more subtle issues you overlooked at first. Your honeymoon period wears off and you see that "the model got dumber". It didn't. You got better at pushing it to its limits, exposing the ways in which it was always dumb.
>Now, will the likes of Anthropic just "API error: overloaded" you on any day of the week that ends in Y? Will they reduce your usage quotas and hope that you don't notice because they never gave you a number anyway? Oh, definitely. But that "they're making the models WORSE" bullshit lives in people's heads way more than in any reality.
Maybe countries could tackle such problems twofold:
- first, implement a nationwide social freezing program, where women in their 20s are offered to freeze their eggs at a young age for free. Such a large-scale program would probably also improve the tech and might make egg collection less intrusive.
- combined with this program, let the women who freeze their eggs opt-in into an egg donation program, where some of their eggs can be used by women with fertility problems
But as with many things fertility, seems that modern states simply do not have the capacity to seriously try anything. Who knows why that is.
They might also look to Israel to see what they're doing that's working so much better than other OECD countries - see my other comment in this post.
But Israel's advantage seems to be partly cultural and I don't see any time-limited elected government willing to expend that much effort to change their nation's culture.
From what I've read, the immediate effect will likely be worse for CO2 emissions, because the alternative to (liquefied) gas is often coal power. Also, the various inputs that are needed for global manufacturing are also affected, so maybe even renewable tech gets more expensive.
I'm not saying that the dependence on the middle east was good, but I think it's good to keep in mind that this was a pretty stable equilibrium even with the various questionable countries involved until the US initiated a global supply shock without a good reason.
There are short term and long term effects. Overall these are good changes.
There are a couple of points to make here. The lead time for new coal/gas plants is years. If it's not planned already, any newly planned plants are unlikely to come online this decade. The supply chains simply can't handle building more turbines and it takes years to fix that. Also, that investment is super risky in it self.
Another point is that the cheapest and fastest way to add new capacity to grids is via renewables. That's why we see record breaking new capacity coming online on a regular basis.
There is indeed a short term increase in emissions from electricity plants because the fastest way to bring more capacity online is to use existing underused plants. A lot of gas and coal plants are no longer running full time because they are too expensive to operate. But they haven't been decommissioned either. Some gas plants actually are used as peaker plants. Most older coal plants take too long to warm up for this. So, yes short term the expensive but quick way to provide extra power is via these plants. But of course, as soon as something more affordable comes online, these things go back to being utilized less. There are many tens/hundreds of GW of renewables and batteries being deployed in the next few years.
Data centers add to all this pressure. That's long term a good thing because these too will want to long term reduce their OpEx by cutting as much dependence on gas/coal as possible.
A final point to make is that despite all these increased emissions, there are also decreased emissions from electrification. Even if the power for an EV comes from an efficient gas/coal plant, it's actually better than the alternative of burning petrol in a combustion engine instead. Less emissions this way. Same for heat pumps. With a COP of 3-4, they outperform burning gas by 3-4x using electricity. Even if that electricity comes from a gas plant operating at 40-50% efficiency. Less gas gets burned.
So these are all good effects even if the reason is a bit sad and unnecessary. This crisis is unnecessary. But I like that it is helping to kill fossil fuel companies faster. This long term erodes confidence in the market as a whole and drives decision makers to do exactly what the article suggests: cutting the dependency on fossil fuels as fast as possible. It's already resulting in measurable reductions in oil/gas imports in some countries.
> An expensive AI which simply takes your job or forces you to work harder
But this implies higher productivity, no? This must mean more outputs that should benefit someone, unless the jobs that are being automated had little value to begin with. Seems paradoxical.
I think there is still the debate to be had whether prediction market enable too much criminal activity and insider trading compared to traditional stock markets and therefore need to be limited for pragmatic reasons (i.e. the legal system can't keep up), but that's a different discussion.
reply