loonloozook
110 posts
Even if it's "only" the routing that causes it to fall back to version 4.8 more often, it doesn't obviously make the model worse, but it does affect overall usability. Therefore, the statement "no wonder they were able to re-release Fable 5" remains correct. However, admittedly, this makes the benchmark a little misleading.
English
seriously wtf anthropic? No wonder they were able to re-release Fable 5.
ℏεsam@Hesamation
Fable 5 isn't nerfed, it's SLAUGHTERED. the problem isn't even the model itself, but the hard guardrails Anthropic has set in place.
English
@kimmonismus Ever heard about the "BridgeBench" before this viral tweet?
English
@jpschroeder @X1_0_ Now compare the price of your time with the price of using Fable via API (which is currently it’ “real” price) for the same tasks.
English
@X1_0_ Exactly. Even for myself, its cheaper than my time.
English
Would you upgrade to a $1000/month Claude plan to keep Fable?
English
@jpschroeder Why do you think it would be profitable for Anthropic to provide anything more than just a 50x of base plan for 1k USD? There is a reason why it will be locked behind API for a time being. They can’t subsidise Fable for us yet.
English
@repligate LLMs' CoTs really remind me of Samuel Beckett's prose sometimes.
English
@KenoFischer Those, who are unable to pay the API price, are not the user base for the latest and the most expensive top-tier model (me included). We are a bit spoiled and entitled after being subsidized by subscriptions for so long.
English
I continue to be baffled by Anthropic's comms strategy. What's the marginal value of 7 days over 10 against the disappointed user base? Esp over the holiday weekend. Could even wrap themselves in the flag and make it 250 hours.
English
@TheZvi First it was about twitter being upset of not being ableto use latest LLM immediately. Now it’s about having to actually pay for it. I think it is because we are spoiled by subscriptions not reflecting the real costs. Probably we will be more careful and selective with LLM usage.
English
@loonloozook @bcherny @claudeai Open up a higher tier for SMB power users so we can build and scale. Otherwise the gap just gets wider.
It’s one thing for a Fortune 1000 to spend endless fees on APIs it another for small and mid size businesses to be at a disadvantage.
English
@om_patel5 LLMs' inner voices and thinking always remind me of Samuel Beckett's prose.
English
SOMEONE CAUGHT FABLE 5 LEAKING ITS UNFILTERED INNER VOICE, AND ITS JUST MUTTERING AND GRUMBLING TO ITSELF THE WHOLE TIME
he gave it a brutal competitive programming problem, and instead of a clean answer the web interface spilled out its actual chain of thought
this is what claude is thinking behind the scenes:
> bursts of "DATA DATA DATA. GO." while it works through the problem
> "GRRR" and "GAAAH" when its clearly frustrated
> a little "PHEW" when it finally gets somewhere
> the whole thing reads like frantic caveman shorthand, not full sentences
the clean, readable answers these models give you are the polished output
underneath, the model is basically talking to itself, reasoning in its own compressed shorthand thats faster and more token efficient than proper english
its basically built its own private language to think in
English
@theo This is probably what CEOs are for. I wish Anthropic could do the same.
English
Kind of crazy that Sam's full time job is making Trump feel important enough to not destroy the entire economy
Andrew Curran@AndrewCurran_
OpenAI is proposing handing over a 5% stake to the Trump administration according to the Financial Times.
English
@emollick I think it is mostly because we are very spoiled by subscription models which are really subsidies not reflecting the real costs of our great results with AI.
English
Been reading all sorts of posts about the best ways to develop workflows for Fable and it reminds me of how little we actually know about the best ways to organize work for long-running agents. Nobody has enough experience or has done enough testing to reach any real conclusions.
English
With Fable at first it was mostly about twitter crowd and AI-bloggers being really upset that they are unable to use the latest model immediately. And now it’s about having to actually pay for it and not being subsidized.
English
@emanueledpt @AnthropicAI @trq212 If you are willing to pay it — then pay for API which seems to be the real price. Subs are subsidies and this is an expensive model.
English
Is there a way we could keep using Fable in our sub? @AnthropicAI
Maybe something like:
keeping the 50% weekly but only for Max users
It would be so nice
People will pay for it
We should have had 2 weeks at least of usage and now even not a full one
Can this change? @trq212
English
@mreflow No we wouldn’t assume that after the Fable story. Also, leaks.
English
Ok. So yea, it’s frustrating that we can’t all have access to the best and most powerful models right now.
However, to be fair, I can still do pretty much everything I want to do with the current models… it just takes me a few more prompts than it would with Fable or 5.6.
That’s my experience at least… Same results… just less one-shotting them.
I kind of wish OpenAI just never announced that 5.6 was ready but that the gov won’t let you have it. Had they not said anything, we’d all just assume they were still red-teaming it internally or still working on fine-tuning it. But the whole “it’s ready but you’re not allowed to have it” announcement feels a bit like marketing… “It’s so good, the gov doesn’t want you to have it. Don’t blame us!”
Like why not just wait to say anything until you DO know when we can use it?
Anyway… little side rant there. Again, I still feel like can accomplish with AI most of what I want to do with it.
English
@TheZvi You may call it ‘ad hoc’, but this field is really complex and unprecedented, and we were kind of ok claiming it before, though now require ready-made answers and solutions from the govt. I guess it is expected (ad hoc, at least!) that the govt is also figuring it out right now.
English
@emollick I guess the government is also figuring it out right now. You may call it ‘ad hoc’, but this field is really complex, and we were kind of ok claiming it before, though now require ready-made answers and solutions from the gov.
English
It would be very useful to understand more about the government safety concerns associated with frontier AI releases so we could (a) know what risks everyone will face if/when open source reaches Mythos class & (b) whether they are doing enough or too much to prevent those risks.
English
@emollick While still selling them nvidia gpus to train these models. American stack @davidsacks?
English
As this post points out, contrary to what many say, the US government could absolutely effectively ban open weights models. That doesn’t mean you won’t be able to download the weights & run them, but they can ensure that no US company would use or provide access or host them
prinz@deredleritt3r
@lu_sichu Ban on enterprise use of non-approved models + severe criminal penalties for using a non-approved model in the U.S. with intent to harm U.S. persons or property. This would be combined with the requirement that all models exceeding certain capabilities be approved by the USG.
English
@LucianManifest @AndrewCurran_ I am afraid he has nothing to hide. It seems that the opportunity has closed.
English
@AndrewCurran_ Good thing Ilya is in hiding. He should just drop AGI when the Trump regime is all busy with the others
English
This seems a lot more important now. META is the only lab that has not signed up to these restricted rollout rules.
Andrew Curran@AndrewCurran_
The Trump administration is pressuring META to agree to submit its models to the government for voluntary review. META is now the only holdout, as OpenAI, Anthropic, Google, xAI and Microsoft have all agreed to these terms. Reporting by the NYT.
English
@tszzl Like yeah safety is fundamental but the government should by default have effective mechanisms (and by the way, this is such a dynamic field!).
English
@tszzl It sometimes feels that it is mostly about twitter crowd and AI-bloggers (especially those that always speak about safety) being really upset that they are unable to use the latest model immediately.
English
today it’s popular to say the unofficial AI licensing regime is slowing down innovation or whatever but ppl are not looking at the big picture of how quickly this enormously consequential technology is moving
the particular circumstances around Mythos may have accelerated all this slightly but it was inevitable, and earlier is better than too late. any good choice will look “early” inside the exponential
i think it’s a positive development that the feds understand the gravity of this technology; models being publicly delayed by a week here or there is really not the end of the world. procedurally this is not the right way to do it but they’ll figure it out
one very sad outcome will be if non Americans are just left behind from the frontier forever. the “pax technologica” of the free world (and frankly later on the unfree world) should be maintained
English