Sign in

David Manheim

@davidmanheim.alter.org.il
4K followers 363 following 1.4K posts

Humanity's future can be amazing - let's make sure it is. Visiting lecturer at the Technion, founder alter.org.il, Superforecaster, Pardee RAND graduate.

PostsRepliesMedia
Reposted by David Manheim
Sanho Tree @sanho.bsky.social · 25/08/2026
“The Trump-installed board that runs the Kennedy Center told a federal judge the building will have to be torn down unless President Donald Trump's name goes on it.“ Even Hitler would have blushed at such an insane demand, but it’s totally normal behavior for Needy Amin.
rawstory.com
Trump board threatens to destroy 'unsafe' Kennedy Center in apocalyptic court filing
The Trump-installed board that runs the Kennedy Center told a federal judge the building will have to be torn down unless President Donald Trump's name goes on it.Justice Department lawyers made the a...
22274105
David Manheim @davidmanheim.alter.org.il · 26/08/2026
I think there are some interesting question at the intersection of LLM personas and agent swarms, and I haven't seen anyone explain this clearly; here is my attempt. davidmanheim.substack.com/p/concerns-a...
davidmanheim.substack.com
Concerns About Personas, Multi-Agent Alignment, and Role Theory
There’s recently been significant discussion of LLM personas and how those fit into alignment.
031
David Manheim @davidmanheim.alter.org.il · 26/08/2026
Public opposition to data centers somewhat matches the vibes of the people worried about AI safety, but actually stopping their construction isn't meaningfully helpful for AI safety, it's at best irrelevant.
040
David Manheim @davidmanheim.alter.org.il · 26/08/2026
@pkrugman.bsky.social Since you're evidently not on twitter:
010
Reposted by David Manheim
David Manheim @davidmanheim.alter.org.il · 20/07/2026
Yes!
112
David Manheim @davidmanheim.alter.org.il · 20/07/2026
I was really happy to appear on the @futureoflife.org podcast this week to talk about @evals-consensus.ai and the challenges of evaluations of AI systems, along with discussions of Goodhart's law, AIxBiosecurity, AI persuasion, forecasting, and human oversight of AI.
youtu.be
Why AI Evaluations Are Broken and How to Fix Them (with David Manheim)
YouTube video by Future of Life Institute
060
David Manheim @davidmanheim.alter.org.il · 20/07/2026
Over on the bird site, Anthropic employees are using their slop machine that 'can't think' to solve longstanding mathematical open questions which have eluded humans for close to a century.
1140
David Manheim @davidmanheim.alter.org.il · 19/07/2026
I keep seeing claims AI companies will get rich, or won't make any money, or that open source models will be cheaper and undercut prices, and similar claims that ignore the simple fact that the economic pricing power is all about compute and token generation costs. davidmanheim.com/AI-Economics/
davidmanheim.com
The Future Economics of LLMs — Interactive Companion
An interactive companion to The Future Economics of LLMs.
160
Reposted by David Manheim
PJ Harry 🎷 @pjharry.bsky.social · 22/05/2026
"Technologies that have first order impacts on coordination and production, or that empower groups in other ways, tend to differentially benefit the powerful in ways that are harmful to others, either directly or indirectly"
021
David Manheim @davidmanheim.alter.org.il · 20/05/2026
New post: Huge technological revolutions aren't usually positive for those living through them - and this bodes poorly for AI even if it is a normal technology. davidmanheim.substack.com/p/if-ai-is-n...
davidmanheim.substack.com
If AI is normal technology, history is not reassuring.
Technological revolutions turn out well eventually, but they go badly first.
070
David Manheim @davidmanheim.alter.org.il · 17/05/2026
As a forecaster, strongly disagree with @kulveit.bsky.social on this. 🧵 "Massive change" is pushing narratives over base rates and trends; in the past 20 years, inequality increased in developed countries. Predictions defying trends based on "this changes everything" are a common error.
110
David Manheim @davidmanheim.alter.org.il · 17/05/2026
GPT-5.5 seems to do more 'hacking per dollar' ...Mythos does more 'hacking per token'" - @peterwildeford.bsky.social Thinking about risks from bad actor's access to scaled models, the costs involved here are still absolutely tiny compared to other attack modes.
030
Reposted by David Manheim
David Roberts @volts.wtf · 12/05/2026
It's hard to avoid the conclusion that Bluesky has been a net negative for US politics. They corralled everyone on the left into a little glass fishbowl where they shout at one another & everyone else ignores them. Meanwhile, all the pols & institutions stayed on X & are being dragged farther right.
12931084152
David Manheim @davidmanheim.alter.org.il · 13/05/2026
The cognitive costs of choosing between different LLM versions or subagents for different tasks, picking or managing the thinking levels, whether to use /fast, and managing token budgets is an amazing illustration of why firms bundling labor into employee salaries is so much more efficient.
000
David Manheim @davidmanheim.alter.org.il · 11/05/2026
I think it's entirely appropriate that the people calling AI a tool are also people who would be insulted if we called them a bunch of tools.
020
David Manheim @davidmanheim.alter.org.il · 11/05/2026
Just sitting here waiting for the @andymasley.bsky.social haters reactions to his newest post...
1211
David Manheim @davidmanheim.alter.org.il · 04/05/2026
Great new piece by a bipartisan team of Ben Buchanan (Biden admin White House Special Advisor for AI) and @deanwb.bsky.social (former Trump admin WH OSTP Senior advisor) saying that AI has national security implications which deserve, but aren't getting, a careful and bipartisan government response.
nytimes.com
Opinion | A.I. Is a National Security Risk. We Aren’t Doing Nearly Enough.
120
David Manheim @davidmanheim.alter.org.il · 04/05/2026
@ioda.live Question: Is February 2021 Texas historical data available? (We keep getting empty responses for historical US bgp data from the API.)
100
David Manheim @davidmanheim.alter.org.il · 03/05/2026
The proportion of Philosophy articles I'm asked to review that have undeclared LLM writing is too damn high! (I think LLM usage for writing is often fine, even / especially if the writing itself is about LLM's ability to reason. But it's supposed to be disclosed, so disclose it!)
040
David Manheim @davidmanheim.alter.org.il · 30/04/2026
"Chinese companies cannot legally fire employees simply to replace them with cost-saving artificial intelligence, courts in the country have ruled, setting a significant precedent for labor rights as automation sweeps the tech sector." There goes the "China is trying to win the AI race" narrative.
caixinglobal.com
Chinese Courts Rule Companies Cannot Fire Workers Simply to Replace Them With AI
Judges classify AI adoption as a controllable business strategy rather than an unavoidable disruption, shielding employees from automation-driven layoffs
000
Reposted by David Manheim
SMBC Comics @smbccomics.bsky.social · 28/04/2026
We’re raising funds to print a brand new book of compiled SMBC comics on the topic of parenting, alongside our preorder for Sawyer Lee! Check out the project here : www.kickstarter.com/projects/wei...
210321
David Manheim @davidmanheim.alter.org.il · 30/04/2026
Hot take from @henryshevlin.bsky.social from over on the bird site, about comparisons of LLMs and humans: "Not a fan of these clichéd “we used to think the mind was clockwork” analogies. Sometimes science just makes progress... Some mechanistic explanations were wrong; others are just true."
120
Reposted by David Manheim
David Manheim @davidmanheim.alter.org.il · 28/04/2026
It's not just in the prompt, it's there twice! (Mistake, or necessary overkill? Who knows!) bsky.app/profile/emol... github.com/openai/codex...
1121
Reposted by David Manheim
Ethan Mollick @emollick.bsky.social · 26/04/2026
339282
David Manheim @davidmanheim.alter.org.il · 23/04/2026
"Draw me a highly detailed where’s Waldo image with people or items to find, but of an ISO SC42 standards conference. Make sure to make it funny with in-crowd jokes."
051
David Manheim @davidmanheim.alter.org.il · 17/04/2026
One problem with making predictions in public is that when I say I'm 75% sure of something, and someone else responds that they are 99% sure I'm wrong, they are using numbers as rhetoric, and I'm trying to make sure that I'd be right approximately 3 out of 4 times.
040
Reposted by David Manheim
niplav is @niplav.site · 16/04/2026
This is the whole alignment problem. All of it, encapsulated. Low-probability behaviors by the model, just make your training environment=your test environment, "what's a capability vs. a propensity", non-adversarial generalization, Goodhart's law. The whole damn thing
121
David Manheim @davidmanheim.alter.org.il · 15/04/2026
LLMs are missing an operating system! Great post by William Waites on the SoTA (Society for Technological Advancement) blog, laying out the argument for what the early history of computing tells us about current and future AI system design. sotaletters.substack.com/p/the-telety...
sotaletters.substack.com
The Teletype of the Future
A brief history of memory: Training data is ROM, the context window is RAM, and tool-accessible storage is disk. What about the OS?
020
Reposted by David Manheim
Ethan Mollick @emollick.bsky.social · 15/04/2026
Comment from a math professor on the quality of the latest proofs.
0557
David Manheim @davidmanheim.alter.org.il · 15/04/2026
Claude refuses to help invent conspiracy theories. Then, after calling Claude a "glorified and electrified rock," the user complains that it's "being emotionally manipulative" and then claims that LLMs will be "used to fine tune human thought as it globalizes our will and ideals."
4284
Reposted by David Manheim
Garrison Lovely @garrisonlovely.bsky.social · 14/04/2026
Sam Altman said he regrets calling the New Yorker profile "incendiary," but never edited his blog post. Now the SF DA is using the same term in a call for deescalation that implicitly frames journalism as a public safety threat. x.com/GerritD/sta...
121
David Manheim @davidmanheim.alter.org.il · 13/04/2026
Slightly contra @davidskrueger.bsky.social's post on the relative merits of AI treaties versus AI regulation, I argue different tools we have should be complementary, and any fights should be about details and different goals, not approaches. Also on LW here: www.lesswrong.com/posts/7CLL4K...
substack.com
Treaties, Regulations, and Research can be Complements
Slightly contra David Kreuger
020
David Manheim @davidmanheim.alter.org.il · 10/04/2026
"Every member of the TESCREAL movement accepts a posthuman eschatology, according to which we should introduce one or more new posthuman species, hopefully in the near future." Did Emile just try to kick me out of the (imagined) TESCREAL community because I don't pass his made-up purity tests?
191
Reposted by David Manheim
Grace @gracekind.net · 09/04/2026
Mythos
512512
Reposted by David Manheim
Cas (Stephen Casper) @scasper.bsky.social · 09/04/2026
🧵🧵🧵 A provocation to the mechanistic interpretability researchers of the world...
1101
David Manheim @davidmanheim.alter.org.il · 31/03/2026
Via CAIDP : "A Los Angeles jury found Meta and YouTube negligent... bolstered a novel legal theory that treats such cases as claims about product design rather than user content." Very exciting to see courts agreeing that design-for-addiction is a litigable harm. www.theguardian.com/media/2026/m...
theguardian.com
Meta and YouTube designed addictive products that harmed young people, jury finds
Jury in Los Angeles awards plaintiff damages of $6m, with Meta to pay 70% and YouTube the remainder
030
Reposted by David Manheim
Asher Elbein @asherelbein.bsky.social · 30/03/2026
It seems like a lot of people have trouble reconciling the fact that while violence *is* a tool, it is also primarily a messy, libidinal impulse that feeds on itself. Many of us have that impulse, and we like to launder it through the righteous causes of others
1284
Reposted by David Manheim
David Manheim @davidmanheim.alter.org.il · 30/03/2026
Twitter is a great place to post about AI as long as you don't care that it's algorithmically down-weighted unless you pay more, you never look at all the bot replies, and you also don't mind the Nazis, personal vitriol, and crypto shilling.
032
Reposted by David Manheim
Ethan Mollick @emollick.bsky.social · 29/03/2026
BlueSky is a great place to post about AI with as long as most people on BlueSky don't notice that you are posting about AI.
1927820
David Manheim @davidmanheim.alter.org.il · 29/03/2026
"We are deeply out of touch with the modern world, and we write like this because the average age of BBC listeners / viewers / readers is over 60, and they finished college before the NES came out."
030
David Manheim @davidmanheim.alter.org.il · 29/03/2026
Embarrassingly wrong. The @quillette.bsky.social article claims AIs: 1. are statistical text remixers, 2. do not reason or understand, 3. are only tools, not agent-like systems. 4. are not conscious. The first 3 are false, it conflates them and lacks a strong argument against #4, which is unclear.
lesswrong.com
Hunting Undead Stochastic Parrots: Finding and Killing the Arguments — LessWrong
I argue the "stochastic parrot" critique of LLMs is philosophically undead—refuted under some interpretations, still valid under others, and persiste…
190
David Manheim @davidmanheim.alter.org.il · 29/03/2026
I'm old enough to remember when my entire shell history wasn't changing directories and calling LLM agents. It was around a year ago, or what is known today as ancient history.
131
Reposted by David Manheim
Alondra Nelson @alondra.bsky.social · 27/03/2026
I’m pleased to stand with The Elders in making this call to action for AI governance.
03510
David Manheim @davidmanheim.alter.org.il · 27/03/2026
I guess I'm mostly back on Bsky now, since twitter is now not only a cesspool, but also again unusable.
Tweets:
Clearly, I'll be less active here now - and not even really by my own choice.

The one thing that made the algorithmically-sorted, politically-biased, ad-heavy, hard to navigate garbage pile of a platform usable was pulled out from under *my already paid subscription* with no notice.

It sure was easy for @elonmusk to convince me not to renew my @X  Pro account - all it took was cutting off a feature which I paid for a year of in the middle of that year, with no warning at all.

(Screenshot saying X Pro is limited only to Premium+ subscribers.)
1141
David Manheim @davidmanheim.alter.org.il · 26/03/2026
I keep seeing people objecting to calls for a treaty addressing AI risk, saying we need details and explanation of exactly what the agreement should do before we ask governments to start negotiating. That's not how the process usually works.
lesswrong.com
"What Exactly Would An International AI Treaty Say?" Is a Bad Objection — LessWrong
I’ve heard a number of people say that it’s unclear what the technical contours of a global AI treaty would look like. That is true - but it’s not ac…
010
David Manheim @davidmanheim.alter.org.il · 23/03/2026
Turns out mainstream sociology is the opposite of science. I'm honestly shocked. It's not like art, which is a valid but different activity - it's actively attempting to undermine collective scientific understanding by banning essential practices among those who might otherwise do useful research.
2223
Reposted by David Manheim
Elad Nehorai @eladn.bsky.social · 13/03/2026
My take on why many (white) progressives say that Israel is making the US go to war is that they would very much prefer that to be the case than to have to examine their own culpability and participation in America's white supremacist militaristic reality. So much easier if it's outsiders.
29754166
Reposted by David Manheim
Alondra Nelson @alondra.bsky.social · 13/03/2026
The Trump admin dismissed AI risk assessment, evaluation and bias testing as burdensome overreach. Now their own DoD is requesting exactly that to ensure systems actually work--in war. Basic, foundational safety and effectiveness testing is essential for *ALL* AI use, as are our rights to privacy.
27635
Reposted by David Manheim
Centre for the Study of Existential Risk @cser.bsky.social · 13/03/2026
A new pre-print "A Labour of Harm: Artificial Intelligence and Biological Weapons Acquisition" discusses how focusing on technological capability risks overlooks the human processes through which biological weapons are imagined Read here⬇️ papers.ssrn.com/sol3/papers....
papers.ssrn.com
A Labour of Harm: Artificial Intelligence and Biological Weapons Acquisition
<p><span>Discussions of artificial intelligence (AI) and biological weapons (BW) have largely focused on laboratory-facing capabilities, treating acquisition as
024
David Manheim @davidmanheim.alter.org.il · 12/03/2026
I expect people on Bluesky will hate this, but... people keep repeating "stochastic parrot" - often without any mental process behind it to specify what the argument is. So I'm writing a paper dissecting various possible arguments, and explaining which are valid. Here's a blog-post version:
lesswrong.com
Hunting Undead Stochastic Parrots: Finding and Killing the Arguments — LessWrong
I argue the "stochastic parrot" critique of LLMs is philosophically undead—refuted under some interpretations, still valid under others, and persiste…
3322