Information persistence hunter extraordinaire. 2x Dad, AI Guy, FP Enjoyer, Math for Fun, Tech Leadership. My views do not represent my employer.
NYC
- Tweets 145,105
- Following 3,137
- Followers 10,008
- Likes 114,129
Rick retweeted
The speed of our current landscape is nuts. Jev was groundbreaking for like 2 weeks.
Cloudflare’s Clef outperforms it, adds vision support and 2x context window.
Oh yeah, and it’s open source.
Cloudflare's Clef beat Jev in just two weeks
How: "frozen" Qwen does one prefill pass, a tiny schema head scores every answer option in parallel, with no text generated at all.
That's why it's 4x faster than Jev at 2x the accuracy on some benchmarks.
And it's open source!
Link to HF:
huggingface.co/Cloudflare/cl…
71
138
3,174
222,170
Rick retweeted
Wake up, baby! Germany has joined the race of frontier LLMs with its sovereign model "Kolibri".
And it's Open-Source, too!
Beats Qwen2635B, Nemotron3 Super, and Mistral Small 4 in the Math, Science, and Code arena! 🎉
The underlying architecture: custom MoE Transformer
Small bird, fast wings, Kolibri is here.
78B parameters. 3.46B active. Up to 1M tokens of context. Built in Europe.
Now the weights are yours. Run it on your own hardware, under Apache 2.0.
74
79
1,667
136,781
Rick retweeted
>hey chatgpt, we don't really need this feature.
>You're right!
+231 -80
>uh... how come removing that feature ended up adding more lines of code
>Good catch! There's a simpler way to implement it
+128 -55
26
25
1,190
27,405
Rick retweeted
This paper shows that AI agents can hit perfect benchmark scores for the wrong reasons, so audit what they did, not just the result.
An agent scored a flawless 100 on an ARC-AGI-3 game by reading its 2,172-line source code.
logs caught what the scores hid.
And then a clean rerun of that game scored only 46.91.
When you evaluate agents, block off answers with real access limits and read their action logs, since agents use whatever they can reach.
18
5
42
3,556
Rick retweeted
I had Claude scroll TikTok, then I gave it 200 bucks and asked for a music video 9 hours later it came back with this... I'm cooked lmaaoooo 😭😭😭
68
71
1,039
86,217
Opus 5.5 can iteratively build insane “slopgarithms” that outperform anything on speed and quality by a large margin. The aesthetics are awful, and would make any human cringe, but it’s hard to argue with the results. Expect many miracle-performing black slopboxes from here.
23
26
1,049
49,315
I’m not saying machines can or cannot be conscious, I haven’t yet heard a great argument either way. I must say though the argument that “it would sure be inconvenient if they were” is completely orthogonal.
1
3
373
Rick retweeted
We’ve confirmed a KVM 0day through our Vercel Sandbox bounty program. Affecting the industry’s gold standard solution for Linux virtualization.
2026 is wild! Thankful to Paulos and other researchers helping us make the most secure sandbox for agents. Full writeup coming.
86
105
1,573
150,493
C++/CLI is one of my favorite programming languages no one really knows about. It’s for .NET and it lets you control managed (gc) and unmanaged (native++) objects independently and differentiate them with syntax. It ends up beautiful except for the normal gross things you get using C++.
6
8
1,237
All I wanted was a coffee, but I bought a breakfast sandwich too because it was only 25% more for the combo.
It makes me wonder if I’m the target here. Margins would have been much higher if I just bought the $5 coffee. I feel like the deal must exist to get people in the door and I’m an unfortunate side effect.
3
3
753
Rick retweeted
And so it continues... Our economic debates seem utterly disconnected from the reality.
That reality: This is an economy mostly creating healthcare jobs, and it does so month after month, and everything else is just noise.
ALT Line chart titled "Effectively all of the net job creation has been in healthcare and social assistance, and jobs are shrinking in the rest of the economy" showing change in non-farm payrolls since January 2025 through September 2026. Healthcare and social assistance added 998,000 jobs with consistent growth throughout the period, while the rest of the economy lost 222,000 jobs, meaning all net job creation came from healthcare. The rest of the economy showed particular weakness from May through August 2025 when losses reached 218,000 jobs. This striking divergence reveals an unusual concentration where one sector's growth of roughly 48,000 jobs per month has offset persistent contraction across all other industries combined.
77
752
2,254
149,324
We drove by a restored house with tile around the entryway, and I remarked to my wife that it makes their house look like a bathroom. Somehow this injured my wife deeply.
1
11
615
I wonder if this makes the final model easier to jailbreak, you’d make it think it’s getting an expert solution. Random and probably naive thought.
SFT is not dead! 🥳
We found a way to make SFT rival current prevailing posttraining methods, often generalizing better and forgetting less than RL and OPSD. 🤯
Following our prior work on reasoning with sampling, we now introduce sampling to the posttraining stack.
1/n
1
487
Rick retweeted
SFT is not dead! 🥳
We found a way to make SFT rival current prevailing posttraining methods, often generalizing better and forgetting less than RL and OPSD. 🤯
Following our prior work on reasoning with sampling, we now introduce sampling to the posttraining stack.
1/n
61
220
1,947
281,507
