tap on the profile, you ain't ready for the description of the hero serving abliterated models
We aim to offer the best models for the most important use cases, as reliably as possible.
In late 2025, we saw the issue with refusals and how terribly restrictive they were for good actors. The bad guys jailbroke models and hosted their own abliterated models while defenders just dealt with refusals. It was immediately clear this wasn't a future that would last. We went against the entire industry’s ideology that centralized access and control to a few labs and their friends was a good idea.
The profound impact of the internet was decentralization of information and control (which rattled the powers that be at the time). We believe that, within the limits of the law (if you break the law, you should be punished), people should have a level playing field in freedom and access. Privacy is a human right, and anything anti-freedom is anti-American.
0
0
0
12
I really really need to know who tf is running these models on max/ultra - I'm sincerely looking for an answer because in the, albeit small, circles of people I see using AI outside of basic use, not once have I seen anyone run max, RARELY xhigh, almost always high
Agent Arena is one of the benchmarks I actually am starting to trust bcz it is built on real people using these models for such tasks
Kimi K3.1 Pro is #11 there, below even Opus 5 and Fable 5 from last gen, which kind of says it is benchmaxxed in terms of claiming that it is comparable with Opus 5.5 x.com/arena/status/2…
1
0
1
41
Claude Sonnet 5.5 (1h 26m): 45.09B in/654.94M out Claude Opus 5.5 (1h 25m): 31.96B in/499.75M out . Gemini 4 Argon (44m 58s): 13.36B in/146.17M out
Beam 1.0 (7m 27s): 4.26B in/44.44M out
absurd.
Beam 1.0 (7m 27s): 4.26B in/44.44M out
absurd.
Reflection's Beam 1.0 now ranks #10 among all open-weight models, is the fastest model on the Vals Index, and is free to the public.
0
0
0
36
joseph retweeted
Just watched the @netflix Instadocs about the @OpenAI cyberattack on @huggingface. Super disappointing, as they went for maximum anthropomorphism of the technology and full doomer mode.
I spent more than two hours with them trying to correct all their mistakes and ground this in t agents are technological systems, not aliens. But it looks like the drive to scare people for maximum views was too strong.
Also, they didn’t mention the lack of containment and monitoring of the agents, how closed-model safeguards blocked our defensive work, or how we defended ourselves with AI using an open model. And that we still lack a lot of information and transparency, including full agent traces and that companies need to remain accountable for what their agents do.
It ends up being a fun sci-fi movie though!
I spent more than two hours with them trying to correct all their mistakes and ground this in t agents are technological systems, not aliens. But it looks like the drive to scare people for maximum views was too strong.
Also, they didn’t mention the lack of containment and monitoring of the agents, how closed-model safeguards blocked our defensive work, or how we defended ourselves with AI using an open model. And that we still lack a lot of information and transparency, including full agent traces and that companies need to remain accountable for what their agents do.
It ends up being a fun sci-fi movie though!
247
342
2,313
191,284
joseph retweeted
1/ muse spark open is the most capable US open-weight model to date.
816B total params, 42B active. 1M-token context. natively multimodal. built for coding and agentic work like long-horizon tasks, hard software engineering, multi-step reasoning.
america! 🇺🇲
816B total params, 42B active. 1M-token context. natively multimodal. built for coding and agentic work like long-horizon tasks, hard software engineering, multi-step reasoning.
america! 🇺🇲
299
461
5,598
685,058
joseph retweeted
Introducing Bee Frontier and Bee Eclipse, two personal AI devices designed to remember the moments that matter most. One wears on your wrist. One attaches to your phone. Both keep you present. ↓
54
37
337
140,883
joseph retweeted
Claude: "omg you put your api key in chat, that's very unsafe, please rotate immediately!"
Elon: "bro just take a photo of your credit card and send it to Grok in chat"
Elon: "bro just take a photo of your credit card and send it to Grok in chat"
To buy stuff, you can just take a picture of your credit card, drop it in the chat and Grok @bot will scour the Internet for the best deal on the item you want and order itCommunity notexAI explicitly warns users not to include personal information in prompts, yet credit card photos may contain such information.
Grok Bot's terms state that user data and credentials may be stored in its cloud environment and require human review for financial transactions.
x.ai/legal/privacy-…
x.ai/legal/grok-bot…
76
163
4,058
159,435
agent orchestration or "make the expensive model the orchestrator" is my #1 bitter lesson.
This is f*cking insane. You've been paying Opus 5.5 to check whether a file exists.
run Opus 5.5, Sonnet 5.5 and Fable 5.1 together, and stop burning Opus on work it was never needed for.
the whole idea in one line:
the strong model plans, the mid-tier model builds, Fable only actually matters.
roles, broken down:
Opus 5.5, high effort, owns the plan and ships the final code
Sonnet 5.5, medium effort, splits into explorer (reads the codebase), worker (edits files, runs tests), researcher (pulls docs)
Fable 5.1, set with /advisor fable, never writes code. Claude calls it at the moments that matter and hands it the whole session
three moments where Fable gets called:
→ a plan goes out: is this actually the right call?
→ the same failure shows up again: is this going nowhere?
→ the task gets marked done: did something get skipped?
Jev engineering does the same thing one level down. the forks that don't need real thought, which file, which tool, keep going or stop, go to Jev and come back in under half a second. the big models only see the forks that genuinely need a decision.
anyone still running one model for everything is paying Opus rates for yes-or-no questions.
drop this into Claude Code:
"Rebuild my Claude Code setup around this structure:
Look through ~/.claude/agents and .claude/agents for subagents already covering explorer, worker and researcher. Only create new ones for roles that are missing. Set model: sonnet and effort: medium on each. If an existing subagent is locked to a different model, leave it as is and just list it.
In ~/.claude/settings.json, set advisorModel to fable. Opus 5.5 ignores a top-level effortLevel there, so remind me to run /effort high while I'm on Opus 5.5.
Check for anything that turns the advisor off: CLAUDE_CODE_DISABLE_ADVISOR_TOOL, DISABLE_TELEMETRY or anything else blocking feature-flag fetches. Report what you find. Don't change any of it yet.
Add one line to ~/.claude/CLAUDE.md: consult the advisor before a big plan, when the same error shows up twice, and before marking a long task done.
Show every change as a diff first. Wait for my go-ahead before touching anything."
0
0
1
31
my favorite trope of demos lately:
"I work in tech, I fly a lot, book hotels, grab dinner with friends, have no kids, so here's how I use AI"
if my mother saw an agent convert a photo of her hand written calendar and put it in her phone's calendar, she'd lose her damn mind.
"I work in tech, I fly a lot, book hotels, grab dinner with friends, have no kids, so here's how I use AI"
if my mother saw an agent convert a photo of her hand written calendar and put it in her phone's calendar, she'd lose her damn mind.
there are so many everyday demos that could make ai, personal agents, & agents in general feel insanely powerful to normal ppl & yet ai companies keep choosing things like booking flights, planning travel, or god forbid planning a god damn wedding.
these are not everyday problems. they’re relatively rare events & worse for a lot of ppl they’re actually part of the joys of life. the best demo should make someone feel pain they already are familiar with disappearing. almost every company gets this wrong. they optimize for spectacle instead of relief which is fine in some cases but it doesn’t last.
one of my favorite historical examples is a product called head on.
the ad showed someone with a headache, then showed them applying the product directly to their forehead while repeating:
“head on. apply directly to the forehead.”
beautifully simple. you instantly understood the product, the problem, & the value. ai companies should be doing the same thing.
show me the annoying thing i deal with every single day then make it disappear. it’s not sexy, i guarantee it will work. when jobs demo’ed the iphone he picked universal simple problems that the iphone did better than anything else on the fucking planet.
anyway, thanks for coming to my ted talk.
0
0
0
47
joseph retweeted
We are introducing Clef and Clef-flash, open-source decision models hosted on Workers AI for high-speed classification and agentic workflows. cfl.re/3Vlqb9P #BirthdayWeek
69
122
837
207,510
joseph retweeted
Nvidia CEO Jensen Huang on AI model distillation:
“That’s called competition. You’re allowed to test somebody else’s products all you want.”
“That’s called competition. You’re allowed to test somebody else’s products all you want.”
100
412
5,477
1,490,187
"Models with these levels of capabilities can also be used by defenders."
>Oh cool, they redeemed themselves from an otherwise terrible blog post
"We're working to give more people, who WE deem worthy, access to OUR models, the safe models, the ones with their CoT hidden"
smh
>Oh cool, they redeemed themselves from an otherwise terrible blog post
"We're working to give more people, who WE deem worthy, access to OUR models, the safe models, the ones with their CoT hidden"
smh
0
0
0
12
joseph retweeted
🚀 Introducing XOR
An open source, multimodal Jev-like decision model. Being developed with data sovereignty & enterprise grade in mind.
❶ Multimodal: great for text + visual classification
❷ Much bigger 260k context window
❸ Qwen3.6 based
huggingface.co/juspay/xor
↓↓↓ See it in action
An open source, multimodal Jev-like decision model. Being developed with data sovereignty & enterprise grade in mind.
❶ Multimodal: great for text + visual classification
❷ Much bigger 260k context window
❸ Qwen3.6 based
huggingface.co/juspay/xor
↓↓↓ See it in action
18
102
675
44,372
joseph retweeted
Just watched a frontier LLM debug a website by literally turning my wifi off and on - but ofc it couldn’t actually turn the wifi back on because its connection was severed
153
195
10,817
219,858
