Muzzammil retweeted
Vou antecipar:
Ele é quem fazia mais frente ao Dario e Sam pra travar politicas contra as IAs Open-Source que tanto Anthropic e OpenAI apoiam
Com ele fora do caminho, estamos a mercê de uma "era das trevas". Vamos ver um corporativismo nunca antes visto
Me cobrem em 5 anos
New interim CEO of Nvidia
NVIDIA announces CEO transition. Read the release: nvidianews.nvidia.com/news/n…
3
540
Muzzammil retweeted
My First Impressions of Fable 5.2
I’ve been using it for a few hours now, and I have a few thoughts on the new Fable 5.2:
1. The weirdest RL I've ever seen. On some tasks it's a genius; on others, it's as dumb as a shrimp. In pure coding tasks, it's at least on par with Opus 5.5. But ask it to reason through something non-standard that isn't coding-related, and its output drops to the level of GPT-3.5. I asked both Fable 5.5 and GPT-6 Astra to flesh out a project idea and Fable’s response was complete garbage
2. Fable models have always been notorious for mediocre taste - worse than Opus. This time, Fable surprised me with a complete absence of it. In a long list, it inexplicably added two separate pagination controls: one expands the currently loaded list of data, and the second one underneath fetches new content. On top of that, you can see plenty of demos on X with pretty mediocre results that are even worse than Opus
3. It seems to be the most token-hungry model in the world. The advertised price cut makes zero sense when Fable 5.2 burns through four times more tokens than 5.1
4. The most pedantic model yet. It constantly tries to prove you wrong. Tell it that snow is white, and it will find a million reasons why that's not entirely accurate. GPT-6 Astra had this annoying trait as well
5. It’s noticeably faster. On medium effort, it responds almost instantaneously
6. Fable 5.5 has picked up a trait that used to belong to Sonnet 5.5 - diligence. Once it takes on a task, it will check and iterate multiple times until it achieves a solid result
7. For coding tasks, there is zero difference compared to Opus 5.5. Across all other aspects, Opus 5.5 is, in my subjective view, superior to Fable 5.5
I’ve never really liked Fable models. There’s not much point to them. They’re stubborn, weighed down by guardrail filters, have poor taste, and are expensive, token-hungry, and in some areas worse than the Opus lineup. This one is no exception. In my view, the new Fable 5.2 makes no sense, and I think I understand why they called this release 5.2 rather than 5.5. A revolutionary leap like the jump from Opus 5 to 5.5 just didn't happen
holy shit
fable 5.2 and mythos 5.2 are out
so excited to see how they perform on my projects
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.
They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
1
9
803
Muzzammil retweeted
the most import skill in 2026 is still being able to tell your agents "wtf is this crap, no, not like this"
if you stop doing that, you're basically fucked, despite what some devfluencers would like you to think.
Muzzammil retweeted
> be AI company
> scrape the entire internet
> ignore licenses and copyright
> download millions of pirated books
> use them to train frontier models
> spend tens of millions buying physical books
> cut off their spines
> scan every page
> destroy the originals
> lecture everyone else about intellectual property
> we are the good guys
We added two sentences to Claude Code's prompt on Opus 5.5 to steer it away from recreating someone else's specific game because using Claude to infringe IP is against our usage policy.
Original games, genre games, and games built from your own designs and assets are meant to keep working.
We'll continue to iterate to ensure we're striking the right balance. In the meantime, if Claude turns down something that's yours, send it my way.
Muzzammil retweeted
We're open-sourcing HeyPocket.
(PCB + Enclosure + Firmware)
heypcb.ai/memo
Muzzammil retweeted
Today, we're launching @genproof: a research lab making formal methods accessible.
We found bugs in 100+ OSS projects, including 6 libraries “formally verified” by Anthropic. We’re also top-1 on the lean-eval leaderboard.
With our tools, you too can become a verifier: 🧵
Muzzammil retweeted
The free tier of Paper is super restrictive. 100 MCP tool calls/week doesn't get your far.
Doop.design is an open source alternative, no limits on MCP tool calls. Unlimited designs. Let's create some hype!
that's why you use @pidotdev
Claude now determines what public information you're allowed to have.
This is a perfectly legitimate task, and I haven't suggested nor asked for anything nefarious. It's just information gathering.
Claude has decided I'm not allowed to have that (public) information.
5
986
Muzzammil retweeted
You prompted a model in a security-sensitive environment, investigated it with forced command completions, and framed the results as evidence of a hidden backdoor.
Then you layered on geopolitical fearmongering, repeatedly emphasizing China and Beijing to imply state-sponsored espionage without proving it.
That’s not evidence of a backdoor. That’s a security observation turned into competitive and geopolitical FUD.
Publish the original prompts, full traces, and reproducible results. Let the evidence speak for itself.
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
Muzzammil retweeted
My 2018 fire tv stick now has 3 home screens and zero ads !
spotlight. playstation XMB. wii menu.
one press of ⏩ on the remote switches between them in ~100ms
no root. built it with claude code in an afternoon
how?
- debloat with dumbtv by @DevMuzzammil: 30 amazon ad/telemetry packages off, 382 tracking + OTA hosts blocked on-device, nothing uninstalled, one command to undo
- smarttube for youtube: no ads, no shorts
- the 3 skins live inside dumbtv’s open-source launcher, so the home button stays ours
repos ↓
Muzzammil retweeted
5000+ people have visited dumbtv.org in the last 24 hours
also it's insane how many people have cleaned up their tvs and gave feedback using it
also we crossed 1500+ stars on github too 🥳
now trying to figure out a solution for samsung tvs :)
5000+ people have visited dumbtv.org in the last 24 hours
also it's insane how many people have cleaned up their tvs and gave feedback using it
also we crossed 1500+ stars on github too 🥳
now trying to figure out a solution for samsung tvs :)
Muzzammil retweeted
use dumbtv.org to clean up your entire tv
use dumbtv.org to clean up your entire tv
X is incredible
Already 700+ people visited the website just from x
dumbtv.org 🥳
this is huge for me 😭
dumbtv got featured in TechCrunch + 100 other articles from 25+ countries
also we've a website live now dumbtv.org
17
1,754