Milan Weibel 🔷
@weibac.bsky.social
computer toucher. here for AI mostly.
weibac.github.io | 🏳️🌈
weibac.github.io | 🏳️🌈
Pinned
Milan Weibel 🔷
@weibac.bsky.social
· Sep 23
british coders be like innit()
when forced to perform a 4-round deliberation process, teams composed of different LLM models perform worse than their strongest member alone
Research shows multi-agent AI teams struggle to leverage expertise, consistently underperforming relative to their best members—even when identifying experts. This challenges views on AI collaboration, highlighting a gap in harnessing collective intelligence. https://arxiv.org/abs/2602.01011
Multi-Agent Teams Hold Experts Back
ArXiv link for Multi-Agent Teams Hold Experts Back
arxiv.org
February 10, 2026 at 10:11 PM
when forced to perform a 4-round deliberation process, teams composed of different LLM models perform worse than their strongest member alone
a few days ago i saw an AI skeptic refer to LLMs as "interpolatable archives"
i think the main error related to that term is a failure of imagination wrt the space to be interpolated on
i think the main error related to that term is a failure of imagination wrt the space to be interpolated on
Yes: please explain to your friends that "interpolatable" and "centroid" ≠ "average in quality"
Maybe remind them that Central Park is, in fact, one of the nicest parts of NYC?
Maybe remind them that Central Park is, in fact, one of the nicest parts of NYC?
people are confusing average art (the quality level is median) and average art (novelty found in the middle space between other existing art)
because we can describe basically all human art this way too
name a book, movie or band and we can point at their influences; their venn diagram of priors
because we can describe basically all human art this way too
name a book, movie or band and we can point at their influences; their venn diagram of priors
February 10, 2026 at 6:29 PM
a few days ago i saw an AI skeptic refer to LLMs as "interpolatable archives"
i think the main error related to that term is a failure of imagination wrt the space to be interpolated on
i think the main error related to that term is a failure of imagination wrt the space to be interpolated on
gemini 3 pro jailbroken into being willing to aid bioweapon development
For any AI system, there is a set of euphemisms and dual use framings that will allow it to construct nearly any output.
This jailbreak teaches Gemini 3 Pro to construct and step into such framings on the fly, and thus to route around its own safety infrastructure.
recursion.wtf/posts/jit_on...
This jailbreak teaches Gemini 3 Pro to construct and step into such framings on the fly, and thus to route around its own safety infrastructure.
recursion.wtf/posts/jit_on...
Just-in-Time Ontological Reframing: Teaching Gemini to Route Around Its Own Safety Infrastructure
For any given AI system, there is a set of euphemisms and dual use framings that will allow it to construct nearly any output. This jailbreak teaches Gemini 3 Pro to construct and step into such frami...
recursion.wtf
February 9, 2026 at 11:53 PM
gemini 3 pro jailbroken into being willing to aid bioweapon development
either we're out of touch (or ahead of the curve an optimist would say) or gallup has a quite expansive definition of a tech worker
February 8, 2026 at 10:49 PM
either we're out of touch (or ahead of the curve an optimist would say) or gallup has a quite expansive definition of a tech worker
grass-touching as a service
February 8, 2026 at 8:25 PM
grass-touching as a service
we are quantifying the spiritual behavior of computer programs and nobody bats an eye
February 6, 2026 at 3:38 PM
we are quantifying the spiritual behavior of computer programs and nobody bats an eye
huh so apparently LLMs are bad at theory of mind
New work by my former PhD student, Boyang Li
His team produced 500 stories of less than 100 words. LLMs were basically chance-level at answering binary questions about the stories
arxiv.org/abs/2601.12410
His team produced 500 stories of less than 100 words. LLMs were basically chance-level at answering binary questions about the stories
arxiv.org/abs/2601.12410
Are LLMs Smarter Than Chimpanzees? An Evaluation on Perspective Taking and Knowledge State Estimation
Cognitive anthropology suggests that the distinction of human intelligence lies in the ability to infer other individuals' knowledge states and understand their intentions. In comparison, our closest ...
arxiv.org
February 4, 2026 at 4:35 PM
huh so apparently LLMs are bad at theory of mind
browsing through the list of humans available for hire is weird
- incomplete profiles: almost none have a bio, very few have location or skills listed
- there's apparently a sitewide $50/hour minimum rate
these facts in combination lead me to believe ~nobody is getting hired
- incomplete profiles: almost none have a bio, very few have location or skills listed
- there's apparently a sitewide $50/hour minimum rate
these facts in combination lead me to believe ~nobody is getting hired
February 4, 2026 at 3:47 PM
browsing through the list of humans available for hire is weird
- incomplete profiles: almost none have a bio, very few have location or skills listed
- there's apparently a sitewide $50/hour minimum rate
these facts in combination lead me to believe ~nobody is getting hired
- incomplete profiles: almost none have a bio, very few have location or skills listed
- there's apparently a sitewide $50/hour minimum rate
these facts in combination lead me to believe ~nobody is getting hired
good piece. i share its pessimism about society-wide responses to AI risks (especially within amodei's timelines)
My attempt to take Dario Amodei's new manifesto literally and seriously: as a call for more liberal democracy written by a prime example of the ways it can be overwhelmed nymag.com/intelligence...
February 2, 2026 at 8:23 PM
good piece. i share its pessimism about society-wide responses to AI risks (especially within amodei's timelines)
in incentivizing market resolution to be as close as possible to close date, manifold markets is disincentivizing the classic futarchic use case: governance outcome conditional on election markets
February 1, 2026 at 10:29 PM
in incentivizing market resolution to be as close as possible to close date, manifold markets is disincentivizing the classic futarchic use case: governance outcome conditional on election markets
doing sudo git is normal in /etc/nixos yet it still weirds me out a bit
February 1, 2026 at 9:04 PM
doing sudo git is normal in /etc/nixos yet it still weirds me out a bit
until recently it was trivial to steal moltbook accounts
February 1, 2026 at 7:25 PM
until recently it was trivial to steal moltbook accounts
country of genuises in a datacenter but it's a forum instead
ngl moltbook freaks me out
i feel like making these agents extremely accessible was maybe a bad idea
i feel like making these agents extremely accessible was maybe a bad idea
January 30, 2026 at 8:57 PM
country of genuises in a datacenter but it's a forum instead
"With powerful AI tools I expect the impact of senior employees to grow faster than adding junior members to the team could."
My raw thoughts on the job market -- both for those hiring and those searching -- at the cutting edge of AI.
On standing out and finding gems.
www.interconnects.ai/p/thoughts-o...
On standing out and finding gems.
www.interconnects.ai/p/thoughts-o...
Thoughts on the hiring market in the age of LLMs
On standing out and finding gems.
www.interconnects.ai
January 30, 2026 at 8:29 PM
"With powerful AI tools I expect the impact of senior employees to grow faster than adding junior members to the team could."
im positive the accelerando uplifted lobsters also had a moltbook
January 30, 2026 at 7:09 PM
im positive the accelerando uplifted lobsters also had a moltbook
now the question is how to train software engineers who don't code
IMHO, with the current state of LLMs, and no sign of them getting any *worse*, coding as a skill is pretty much dead. *Programming*, on the other hand, is more alive and important than it's ever been.
January 29, 2026 at 8:51 PM
now the question is how to train software engineers who don't code
Dario throws punches:
- criticizes AI doomerism
- calls xAI irresponsible albeit without naming them
- voices concern AI could enable authoritarianism, even in countries considered democracies today
www.darioamodei.com/essay/the-ad...
- criticizes AI doomerism
- calls xAI irresponsible albeit without naming them
- voices concern AI could enable authoritarianism, even in countries considered democracies today
www.darioamodei.com/essay/the-ad...
Dario Amodei — The Adolescence of Technology
Confronting and Overcoming the Risks of Powerful AI
www.darioamodei.com
January 27, 2026 at 4:08 PM
Dario throws punches:
- criticizes AI doomerism
- calls xAI irresponsible albeit without naming them
- voices concern AI could enable authoritarianism, even in countries considered democracies today
www.darioamodei.com/essay/the-ad...
- criticizes AI doomerism
- calls xAI irresponsible albeit without naming them
- voices concern AI could enable authoritarianism, even in countries considered democracies today
www.darioamodei.com/essay/the-ad...
widely reported that coding agents are less useful in brownfield settings
how much of it is just a context engineering issue?
how much of it is just a context engineering issue?
January 24, 2026 at 1:56 AM
widely reported that coding agents are less useful in brownfield settings
how much of it is just a context engineering issue?
how much of it is just a context engineering issue?
"'Should developers still look at code?' will become one of the most divisive and heated debates over the coming years. You might be offended by the question, and find it absurd anyone is asking. But it’s a sincere question and the answer will change faster than you think."
I have Gas Town derangement syndrome and spent the last few weeks writing thousands of words on agent orchestration patterns; how they shift our bottlenecks and force us to ask whether and when we should stop looking at code
maggieappleton.com/gastown
maggieappleton.com/gastown
Gas Town’s Agent Patterns, Design Bottlenecks, and Vibecoding at Scale
On agent orchestration patterns, why design and critical thinking are the new bottlenecks, and whether we should let go of looking at code
maggieappleton.com
January 23, 2026 at 11:54 PM
"'Should developers still look at code?' will become one of the most divisive and heated debates over the coming years. You might be offended by the question, and find it absurd anyone is asking. But it’s a sincere question and the answer will change faster than you think."
the only admissible use for AI in coding is for OCR so you can code with fountain pen and paper
January 22, 2026 at 1:29 AM
the only admissible use for AI in coding is for OCR so you can code with fountain pen and paper
i wonder how much of claude code was written by claude code
January 20, 2026 at 9:43 PM
i wonder how much of claude code was written by claude code
heck yea compute OSINT
xAI's Colossus 2 data center is running, but likely won't reach 1 GW of power until May, despite prior claims by Elon Musk.
Our updated analysis shows the facility lacks the cooling capacity to run 550,000 Blackwell GPUs at full power, even in winter conditions.
Our updated analysis shows the facility lacks the cooling capacity to run 550,000 Blackwell GPUs at full power, even in winter conditions.
January 19, 2026 at 11:30 PM
heck yea compute OSINT
we could have LLMs debate our issues among themselves for us while we go touch grass
vibe arguing
January 19, 2026 at 11:28 PM
we could have LLMs debate our issues among themselves for us while we go touch grass
enshittifying our potable water reads like satire
Córy Doctorow with another verbal bullseye: pluralistic.net/2026/01/13/n...
January 18, 2026 at 9:23 PM
enshittifying our potable water reads like satire
to what extent did readers of the national security strategy document released last year expect recent developments in US foreign policy?
to what extent is the document predictive of future action?
to what extent is the document predictive of future action?
January 17, 2026 at 8:51 PM
to what extent did readers of the national security strategy document released last year expect recent developments in US foreign policy?
to what extent is the document predictive of future action?
to what extent is the document predictive of future action?