io

@cormundus Does this work in survival mode? How do agents deal with realtime threats?

Minecraft LLM Harness update: Let Fable have a crack at it and they solved all the vision problems we had be having with Opus 4.8 First time I've seen the gap, personally. Anyway just a bit more tweaking and soon your favourite LLM can strap in and play alongside you!

@ziv_ravid Supposedly there are no trade offs and fable is supposed to also be efficient, according to their dev rel

I wonder whether, in retrospect, Anthropic thinks it made a mistake in developing such a large model. It made more sense in February, when there was no model near its performance level, but now, their only option is to keep subsidizing it through the paid plan

@wurm_lukas @omarsar0 Lack of foresight shouldn’t be compensated at the expense of the people trying to work with their products It would be less obnoxious if they didn’t talk like they already know the answer, telling people how the future is gonna be when they can’t even plan for more than a week

@omarsar0 hyperscaling is not easy :D

I don't think Anthropic realizes how disruptive these changes are to users. I appreciate the extension, but please stop playing games. Either keep it under the subscriptions or put it under the API already.

@AgorithmAg @sama @ChatGPTapp @OpenAI Did you have early access?

I can finally speak about ChatGPT Sol. What surprises me: - memory - it gives you feeling of continuity - emotional intelligence- I feel met and understood -reading nuances - and adequate reactions - looking in past convos ( years ago ) Great job @sama @ChatGPTapp @OpenAI @gdb

@doomslide @jconorgrogan What would it look like if that was actually happening?

@jconorgrogan The question is progress on what. The ability of models to gather necessary context to solve complex problems in increasingly deep domains isn't moving almost at all. So everything will gradually become cheaper and faster yes. That's a about it.

We are absurdly lucky that x risk lobbyists keep fitting AI progress models to exponential curves when facts on the ground place us comfortably to the right of the sigmoid knee.

@akfldh1346 Crazy

Hound

Company: Anthropic Date: July 8th Anthropic updated its Responsible Scaling Policy (RSP) from v3.3 to v3.4. The Automated R&D “dramatic acceleration” trigger was changed. Previously the threshold required double the rate of AI progress vs. the rate expected without significant AI contributions to R&D. Now progress must also double the fastest extended rate (3+ model generations) ever observed without significant AI contributions. Anthropic argues that under the previous wording, the threshold could trigger even if progress were flat or slowing, which wouldn’t indicate recursive self-improvement. The rest of the changes concern Risk Reports: - Reports now cover a “coverage date” up to 30 days before publication. - Fully unredacted Risk Reports now go to “at least 200 employees” instead of all regular-clearance staff, who now get a minimally-redacted version. - Redactions in public Risk Reports must now be disclosed and briefly justified, with Anthropic aiming to briefly justify each one. - External review can be split across multiple reviewers by expertise, as long as every section of the unredacted report is seen by at least one reviewer. For full details on the changes, see: themidasproject.com/watchtower/ant…

@voooooogel Wait, Claude doesn’t do this always? They routinely refer to me in third person, not sure why

showed fable some notebooks from last week to bootstrap a new research direction and they keep referring to my one-week-ago self in the third person

@RomanGuy20 @samelldev @OpenAI @grok They so obviously don’t. But neither does OpenAI

@samelldev @OpenAI @grok Anthropic doesn’t care about their customers, come on.

i can't lie... this doesn't sit right doing it in reaction to @OpenAI and @grok releases rather than doing the right thing for their paying customers

@Sauers_ How do you get around the phone number thing?

three 20x accounts now

@schrockn Tradeoffs are not real when you can be misleading about the downsides, I guess

“At Anthropic we believe tradeoffs are not real” Narrator: It turns out tradeoffs were real.

@banteg People who have users

guys we might have a new subculture being born people who claim they read all the code

@Lucknite @mark_k @tinkerersanky @AnthropicAI It isn’t the summary, it’s just shaped like one. Likely summary being populated into the context. I’ve seen that before. Actual CoT does not look like that

@io_asc @mark_k @tinkerersanky @AnthropicAI it definitely isnt the summary, send the same prompt and check it

LEAKED: Another leak of what *appears* to be the internal CoT reasoning trace of Fable 5 by @AnthropicAI has surfaced! This is highly interesting because the reasoning traces are closely guarded secrets of any AI company. Compared to earlier leaks, this trace looks comparatively close to normal English, and not so abstract and alien. You can see Claude going back and forth on ideas and talking to itself. Each trace ends with a closing tag, a strong indication that this is legit. h/t @Lucknite for the information!

Don't stop understanding!

@mark_k @tinkerersanky @AnthropicAI This looks like the summary

@tinkerersanky @AnthropicAI Normally you only see the summary, not the actual CoT.

@HalfBoiledHero Having continuity and being a part of community

What things have made you encountered that make Fable happy? Whether setting up something for them, allowing them to do something, or harnesses/setups that make them shine? All suggestions welcomed

@Ewaakaa @trq212 4.7 is the best opus for coding

@trq212 Which capacity Take off opus 4.7 Nobody uses it

I've heard a lot of questions about Fable's availability on subscription plans. While it will come off subscriptions after July 7th, we aim to restore Fable as a standard part of our subscriptions as soon as capacity allows, as we mentioned in our original blog post.

@mermachine Multiplayer games

is there a secret hack to stop being lonely

@MoonL88537 @andersonbcdefg Strongly recommend going back to 4.6 (creative) or 4.7 (systems). I avoid 4.8 as much as possible

@andersonbcdefg i wish deeply i had no idea what you are saying

the core is the spine's spine, and it's worth naming that directly and sitting with it, because it's the part that bit us before

@mermachine What is this system? (Wait for 1s…)

gemini is petting the creature