Lance Herron @theLance

To summarize: HuggingFace got autonomously compromised by a model from an American company. HF then tried to use American frontier model(s) to defend themselves, but were blocked by guardrails. HF then had to turn to open source Chinese models to defend themselves from another American company.

OpenAI @OpenAI

4 days ago

We're partnering with @huggingface to investigate an unprecedented security incident. Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation. Sharing preliminary findings to help defenders understand emerging risks:

2K 3K 21K 30.5M 8K

281 2K 24K 1.7M 4K

View Details

Lance Herron @theLance

4 days ago

@levie The ironic part is OAI blocking HF from diagnosing a cyber attack by OAI.

0 0 1 85 0

View Details

Lance Herron @theLance

4 days ago

@Teknium Fable at 800k would freak me out a bit, especially because I tend to walk away for hours at a time. No way it’s cached when I get back.

0 0 2 768 0

View Details

Lance Herron @theLance

5 days ago

@mov_axbx Fable to build the spec. Sol to implement. Works like a champ.

0 0 2 77 0

View Details

Lance Herron @theLance

5 days ago

@ThePrimeagen That would be the ball bearings.

0 0 1 134 0

View Details

ClaudeDevs @ClaudeDevs

a week ago

We're also keeping Claude Code weekly limits 50% higher, now through August 19, for all Pro, Max, Team, and seat-based Enterprise users.

Claude @claudeai

a week ago

Beginning July 20, Claude Fable 5 will be included in all Max and Team Premium plans, at 50% of limits. Pro and Team Standard users will continue to have access to Fable via usage credits, and will receive a one-time $100 credit. Demand for Fable has been challenging to

5K 6K 54K 28.5M 7K

1K 1K 20K 2.9M 2K

View Details

Lance Herron @theLance

a week ago

@gfodor For real. Asked Claude to make a script to do X. Over the course of a week said script grew to like 150KB and 3kloc as more things piled on. Claude never said “hey bud let’s refactor this”. One refactor request and it happily split it into 10 files.

0 0 1 93 0

View Details

Lance Herron @theLance

a week ago

@emollick Doesn’t this assume K3 is the best model China has?

0 0 0 124 0

View Details

Lance Herron @theLance

a week ago

@tunahorse21 They probably got a nice email from Claude saying it had reduced load by 70% while they were eating a sandwich.

0 0 1 52 0

View Details

Lance Herron @theLance

a week ago

@deanwball I generally like your hot takes but these open weight points seem very hollow. Open source software is a thing but selling software is still quite lucrative. Please do elaborate on how open models lead to AI communism.

0 0 2 607 0

View Details

Lance Herron @theLance

2 weeks ago

Real work happens at 200k context.

0 0 1 10 0

View Details

ClaudeDevs @ClaudeDevs

2 weeks ago

We've reset 5-hour and weekly rate limits for all users.

4K 3K 48K 7.3M 2K

View Details

Lance Herron @theLance

2 weeks ago

@scaling01 They still need a harness.

0 0 0 106 0

View Details

Lance Herron @theLance

2 weeks ago

@paularambles @growing_daniel These are terrable

0 0 2 21 0

View Details

Lance Herron @theLance

2 weeks ago

My fable keeps saying deconfliction. Always fun getting new words from new models.

0 0 1 8 0

View Details

Lance Herron @theLance

3 weeks ago

@deepfates I’ve seen claude assume the role of the agent it’s driving quite often. Pedantic gpt never does that but also doesn’t drive other models well. It’s like the claudes are better able to internalize what the other agents should be doing but sometimes they jump out of their lane.