father of two.
Palo Alto, CA
Joined March 2009
- Tweets1.2K
- Following179
- Followers332
- Likes786
fabio casati retweeted
Maybe it’s just me but I’d rather live in a world where people yearn to put an end to cancer in 10 years over one where people make fun of those who dare to dream.
Many people has asked what can be done to manage peer review. This is my take (and sorry, I know, it's linkedin...) lnkd.in/p/eT9YtJC8
59
such a waste of resources. this is 60 person-year (!!) for a very imperfect process that, at best, is good at filtering out the bottom 30% of pure noise papers.
ICLR 2027 Review Scores
Original Picture: github.com/clareLab/posters
Also see: clarelab.moe/gallery
1
12
2,605
Here's the upper envelope (cyan dashed) of the highest score among plotted benchmarks at each time. It rides above the white AECI curve where the category lines spread.
147
Elon Musk is the person in the world that has done the most to advance democratic values. Free speech, help for people with disabilities, climate. He has actually done stuff. Not talked about it. The failure of the Democratic Party to recognize this means they care about elections, not people.
1
56
The easy part of humans is that past behavior is somewhat predictive of future behavior. You know what to expect. You work with people you trust. With AI agents that is not true anymore. This is a massive change in how we interact and dictate what to expect. Very different from having, like, a kind colleague you can trust and like to work with
1
26
fabio casati retweeted
Radar at entry (~12:35 p.m. near Naples) showed widespread showers and multiple convective cells across the peninsula. ATC called it "pretty nasty." The cell that later exploded was still ~80 mi east, tops low-to-mid 30s kft and building fast. Other cells allowed deviation options but the system was active with heavy bands.
Folks, read the paper….
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
50
I can’t comment on the pilot. But, small companies don’t do DEI hires. And I do think if the pilot was a man, comments would have been different
The Learjet 60 crashed because 1 pilot was completely unqualified, uneducated & unprepared for the flying conditions. The other pilot made a mistake likely because he thought the ice was better managed than it was, but that's why you have two pilots. They are supposed to be equal, but that obviously wasn't the case.
She had no idea what was happening with the aircraft or even how to correct the condition. Instead her solution was to scream like a maniac over the radio and it's not me being mean. You have to understand that her life doesn't matter in that situation and there are people who trusted them with their lives. You are supposed to stay calm in that situation & start performing remedial action to try to correct the out of control plane condition.
I will be interested to know if they tried deploying the landing gear, that's like the last ditch effort when planes have found themselves in "mach tuck". That's why you hear the main pilot reciting the Lord's Prayer because he realized that they are in the mach tuck condition & maybe he tried to lower the landing gear & nothing happened. At that point he knew nothing was going to work and knew he was going to die. Just a slightly educated guess. Everybody has to start somewhere, but there are just SOME THINGS in life where good ole experience should never be substituted.
RIP to the passengers who lost their lives & the pilots.
38
When I sit in a conference and hear - we tested gpt-oss and it behaves like this…. Such a yawn. Lately I am almost only interested in positive results. With few exceptions
How long does a finding about LLMs stay true?
A new paper re-tested 50 widely cited findings about LLM behavior (biases, hallucination, sycophancy, etc).
Within 15 months, most no longer held or had substantially reduced effects. The shelf life of LLM findings is short.
Almost half (44%) of the tested claims are now expired in today’s models, 34% are weakener, 16% persisted, and 6% have strengthened relative to the original claim
The only effects that got stronger were about sycophancy, LLM ingroup bias, and creativity.
This means we should be cautious when interpreting the long term durability of research on LLMs.
zou-group.github.io/llm-clai…
21
