Prathmesh Pandey

Prathmesh Pandey@file_mutex

23

David Cramer@zeeg·17h

@0xwilt @dillon_mulroy @file_mutex I created a multi billion dollar infra company. We read code here.

English

4

1

56

1.7K

David Cramer@zeeg·1d

codex writes the most digusting code idk who's responsible for pre-training over there but you gotta flip the script

English

89

14

743

114.1K

Prathmesh Pandey@file_mutex·2h

@ngriffin_uk @zeeg can you outperform fable? x.com/file_mutex/sta…

sota coding harnesses are fundamentally better at reasoning and reviewing code than any human. your belief that your biological brain can out-review them on logic isn't "accountability," it's just ego. reading every single line doesn't guarantee quality, it just adds the most error-prone, high-latency component back into the loop: you.

English

0

103

Nicholas Griffin@ngriffin_uk·1d

@file_mutex @zeeg absolute rubbish 🤣 the level of arrogance and unfounded confidence that ai gives people is insane.

English

0

21

435

Prathmesh Pandey@file_mutex·2h

sota coding harnesses are fundamentally better at reasoning and reviewing code than any human. your belief that your biological brain can out-review them on logic isn't "accountability," it's just ego. reading every single line doesn't guarantee quality, it just adds the most error-prone, high-latency component back into the loop: you.

English

0

212

Dillon Mulroy@dillon_mulroy·16h

i’m not sure what your point is here, i build with ai all day long and build ai focused products at cloudflare, but i’m now where close to being naive enough to think these agents and models and produce quality on their own. i very actively steer, readjust, and direct them to get good out comes and that includes reading every single line they produce. why? because agents don’t remove accountability

English

0

38

533

Prathmesh Pandey@file_mutex·3h

@andrewqu Folks who have even basic instrumentation will note it right away.

English

23

Andrew Qu@andrewqu·12h

Hot take: a lot of people wouldn’t be able to tell the difference if they were randomly routed between gpt-5.5, opus-4.8, or fable-5 for their day to day work

English

305

45

1.6K

97.4K

Prathmesh Pandey@file_mutex·17h

@dillon_mulroy @zeeg will do right after Cloudflare deletes all of the "AI" branding from their website @eastdakota

English

0

774

Dillon Mulroy@dillon_mulroy·18h

@file_mutex @zeeg delete your account

English

0

79

793

Prathmesh Pandey@file_mutex·21h

@kskrygan on my codebase, fable was 10x smarter than opus...

English

1

124

Kirill Skrygan@kskrygan·1d

Real assessment of Fable5 from engineers around me: -somewhat better than Opus on OSS repos -about the same on closed-source repos -much more expensive So for real orgs, the value prop is pretty vague But sure, keep believing it was so powerful the government had to ban it

English

15

5

105

11.8K

Prathmesh Pandey@file_mutex·1d

@mattshumer_ That's the problem with unhardened harnesses if one need to wait for better models.

English

The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Claude models is not affected. We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible. Read our full statement: anthropic.com/news/fable-myt…

1

161

Matt Shumer@mattshumer_·1d

Assuming Anthropic is able to restore Fable in the next few days, there's literally zero point doing any meaningful work until it is back. What can be done in 100 hours with Opus can be done in 1 with Fable. Hopefully this is figured out quickly.

Anthropic@AnthropicAI

English

649

201

5.1K

1M

Prathmesh Pandey@file_mutex·1d

@banteg have you tried looking at the brighter side?

English

The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Claude models is not affected. We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible. Read our full statement: anthropic.com/news/fable-myt…

108

banteg@banteg·1d

wow fuck you anthropic, i pay $200 to use the model for one day? a reset won't help here. completely frauded out.

Anthropic@AnthropicAI

English

27

2

221

16.5K

Prathmesh Pandey@file_mutex·1d

Et tu, $AMZN? Anthropic ditched Google for Amazon, just to have them get cheated on loll

NIK@ns123abc

🚨US government’s action to shut down Anthropic’s top AI models was actually triggered by an unnamed rival company claiming it could break Mythos’s security, not by China

English

742

Prathmesh Pandey@file_mutex·1d

@zeeg hmm there are perhaps five people out there in the world, who can beat _the_ sota coding harness. no one will be able to read and understand that much code. dig in when required? sure. but know it in-n-out? most probably not.

English

14

0

2

4.3K

David Cramer@zeeg·1d

@file_mutex people who dont read the code are not serious people and it takes a serious person to ship production software

English

24

53

593

54.7K

Prathmesh Pandey@file_mutex·1d

@davidcrawshaw Looks like you have been out of the '/loop'.

English

116

David Crawshaw@davidcrawshaw·1d

Current status: data analysis and code analysis (and both combined!) with Fable. It appears unmatched at extracting insight from a mountain of code and logs. Then I take the last stanza it creates and hand it to another model for implementation.

English

0

28

2.4K

Prathmesh Pandey@file_mutex·2d

My MCP is roughly saving 50% on blended token consumption in codex and claude. That doesn't mean codex, claude can't build something similar but as a server owner their philosophy will be rooted in "seeing the maximum queries coming through".

Dan Robinson@danrobinson

If you’re proud of your really sophisticated skill or harness, try benchmarking it against a simple one-sentence prompt as a sanity check Codex, Claude Code, and ChatGPT Pro are really, really good

English

321

Prathmesh Pandey@file_mutex·2d

@AndrewCurran_ @bayeslord It def has big model smell. I have my reviewers on GPT 5.5 xhigh, and Opus 4.8 used to stumble through 10 revisions before getting past the reviewers. Fable takes 2-3 revisions.

English

1

232

Andrew Curran@AndrewCurran_·2d

@bayeslord I've been trying to tell people. And Fable isn't the new Mythos. And the new Mythos isn't what they have internally.

English

0

95

4.4K

bayes@bayeslord·2d

Fable is in fact Built Different

English

1

78

6.4K

Prathmesh Pandey@file_mutex·2d

@doodlestein @__paleologo seems reasonable given that he says "pro" -- which gives 20x less usage than max.

English

1

193

Jeffrey Emanuel@doodlestein·2d

@__paleologo This seems to really vary a lot. I’ve been surprised by how much mileage I’ve gotten so far with Fable across a variety of tasks. Granted, I have 22 Max accounts, but it’s not like I’ve blasted through all of them already either. I’m asking them to do really hard stuff, though.

English

6

0

25

5.1K

Gappy (Giuseppe Paleologo)@__paleologo·2d

Clearly, Fable is doing a lot of work, and unleashing a ton of agents. To review a short technical note, it released 31 agents, coded simulations to verify my results, did "adversarial reviews". Eventually, it only made the assumptions slightly more rigorous. It is all good. For a four-page technical note+a little code, though, it consumed all my Pro session tokens, *plus* $17 worth of credits. It is ridiculously expensive. I have 20-page reports that are way more complex than this. I can see how Anthropic has entered the phase of market-clearing prices, yield management, and pre-IPO. I recall Boris Cherny saying in a podcast, "run Opus [4.6], not Sonnet. It's worth it". I feel comfortable saying that running your top-shelf model is *not* worth it anymore. Decreasing returns, on most tasks. Like in the real world, some people can be real smart, but real expensive.

English

58

1.1K

163.2K

Prathmesh Pandey@file_mutex·2d

@claudeai Has Fable not been trained to invoke MCP servers yet? I don't see it doing so. @AnthropicAI

English

37

Claude@claudeai·4d

Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use. Its capabilities exceed those of any model we’ve ever made generally available.

English

5K

14.5K

104.6K

55.7M

Prathmesh Pandey@file_mutex·4d

@jeremyphoward @finbarrtimbers If everyone else uses a frontier model except them, then they won't be the frontier much longer.

English

Jeremy Howard@jeremyphoward

1

59

Jeremy Howard@jeremyphoward·4d

@finbarrtimbers If they believed that, they'd be doing the opposite of what they chose. x.com/jeremyphoward/…

Easy solution to slow down recursive AI self improvement: - The lab with the top-ranked model must agree THEY must not use it for working on frontier AI - But everyone else should have access to it. By definition, this means the frontier doesn't advance.

English

6

1

104

8.5K

finbarr@finbarrtimbers·4d

As my entire feed is criticizing Anthropic, I think that the team there genuinely believes what they’re saying. It’s not a marketing/anticompetitive tactic. They genuinely believe these models are dangerous and that AI research should be slowed down.

English

101

12

406

94.6K

Prathmesh Pandey@file_mutex·4d

Fable has nothing to do with your ability to orchestrate agents.

Ed Zitron@edzitron

There it is

English