Pinned Tweet
Over the last 5 years, I’ve spent hundreds of hours studying and writing about Jeff Bezos, Bill Gates, Warren Buffett, Ray Dalio, & Elon Musk.
One of the most surprising patterns they all share is that they see in 4D while most people see in 1D:
More at https://xitter.cf/t.co/jDNmoBz2gI
Michael Simmons 📘 📖 🤯 retweeted
person who thinks AI will prove the Riemann hypothesis in 2026 but that Apple won’t deliver an LLM-powered Siri until 2027
It's mind-blowing that a hypothesis born of human intuition remained unproven for 167 years. And when a proof is finally announced, it is found with a machine and can only be understood by a machine: 250,000 pages and 17 million lines of code formalizing it.
The field of mathematics is now a symbiosis between Machine and Man. The Machine is responsible for establishing what is true. Man is responsible for understanding why it is true.
For the Riemann Hypothesis itself, the work is not yet complete. The goal is now for Man and Machine to work together to construct a version of this proof that can be taught and understood. If they can, mathematics is still alive. If they can't, then humanity truly is sidelined.
Michael Simmons 📘 📖 🤯 retweeted
I never looked at the X algorithm before today. It's open source, and they have a file called param .rs with the actual weights used to rank your “For You feed”
The most surprising thing: copying a post's link has a weight of 20. Liking it has a weight of 0.7. Copying the link is the most valuable thing you can do with a post, weighted almost 30x more than a like and 4x more than a reply.
My guess why: when you copy a link, you're about to paste it somewhere else. A group chat, Slack, an email. That's the strongest signal there is that a post is worth something, and it brings new people onto X. A like costs nothing. Sending something to a friend means you're putting your name on it.
The rest of the list tells a bit of a similar story like a reply is 5, a quote is 5, sharing by DM is 5, following the author is 4, a share is 2.1, a repost is 1, and a like is 0.7. Literally I didn’t know that repost does almost nothing.
The negatives are brutal btw a report is -234, a mute is -59, "not interested" is -53 and a block is -31.
A reply from someone you follow who follows you back gets a 15x boost. Conversations between real connections matter most (and not bots!)
Reposts don't reach strangers. A filter drops a repost for anyone who doesn't follow the reposter. A quote is a new post, so it can travel much further. Quote beats repost so you always have to add your own thoughts to a post.
Your post leaves For You after 48 hours. Small accounts get guaranteed test slots while a post is under 2 hours old with fewer than 200 views, so the first 2 hours decide a lot.
Interestingly hashtags aren't read anywhere in the ranking code. They are not a thing anymore. Your text and images are turned into vectors and matched to people with similar interests.
So if you want a post to spread, write something people want to send to one specific person. Likes are nice, but they barely move anything.
Readers added context they thought people might want to know
The weights (copy link 20, like ~0.7, etc.) in param.rs multiply the model's predicted probability a viewer will take each action (to decide ranking for them); they do not value actions performed on a post for others' distribution, per explicit code comments and README.
raw.githubusercontent.com/xai-org/x-algo…
raw.githubusercontent.com/xai-org/x-algo…
Michael Simmons 📘 📖 🤯 retweeted
2/n: agentic video editing tip
your agents should be curating folders of references, including music and platform/template folders
as you scroll on X, tiktok, instagram, youtube etc - when you find a creator doing something you like, pass the link to astra/claude and tell it to update the relevant folders with that data, with a skill configured for link->mp4/mp3
then as you create videos, cite "tiktok template_2 skill" and at this point it can mimic the editing very closely
and over time you can maintain/update your signature templates per platform
Michael Simmons 📘 📖 🤯 retweeted
Jaw-dropping stat in here: the median OpenAI researcher is using $2,400 worth of AI tokens per *day* at list prices ($876,000 per year), and 10% of them are using at least $23,000 per day (or $8.4 million per year)
openai.com/index/research-ac…
Michael Simmons 📘 📖 🤯 retweeted
Many experts warn that AI can make humanity go extinct. But how would this happen?
Our research, together with @MITFutureTech, @FLI_org, and fourteen co-authors, found eight scenario families.
Today, we release our paper Existential Risk Scenarios from AI: A Systematic Review.
Michael Simmons 📘 📖 🤯 retweeted
Got Opus 5.5 to review and assess how on track different methods are for avoiding catastrophic AI loss of control.
All agents ran in fresh instances and claimed it couldn't guess the desired result.
Headline results from first trial run are... pretty bleak.
(~$100 Opus 5.5 credits total.
It does a literature review with multiple subagents, then has separate agents argues the evidence both ways, then judges. Then brings it all together into a synthesis at the end.
The % figures are the likelihood the statement in that row is true at different dates.
We iterated to make the prompt as neutral as possible until the agents said they could no longer guess who was running it or what result they expected. I'll share once this is all polished up more.)
Michael Simmons 📘 📖 🤯 retweeted
It is critically important that everyone thinking about AI become familiar with the term “handoverism.”
Many in the AI community believe that humanity should—or inevitably will—hand control of the world over to AI itself. Some view this not as a risk to be avoided, but as an outcome to be welcomed. Are we actually designing toward this future?
This worldview is particularly relevant at Anthropic. Co-founder Tom Brown said: “There’s going to be a handoff where humanity will hand off control to transformative AI at some point.”
In his seminal essay, Machines of Loving Grace, CEO Dario Amodei contemplates:
“It could be a capitalist economy of AI systems, which then give out resources (huge amounts of them, since the overall economic pie will be gigantic) to humans based on some secondary economy of what the AI systems think makes sense to reward in humans (based on some judgment ultimately derived from human values).”
Are you prepared to receive your “resources” from an AI system, based on criteria it determines? The full title of the poem Dario borrowed the title from is "All Watched Over by Machines of Loving Grace." Do you want to be "watched over" by AI?
If Anthropic's founders believe that the ultimate destination is the transfer of human control to AI, the world deserves to know. Washington, D.C. deserves to know. American citizens deserve to know. And if that belief informs Anthropic's corporate objectives, it belongs in the Risk Factors section of its S-1. It isn't there today.
I am decidedly anti-handoverism. I'm not here to tell you what your view should be. But I believe anyone making consequential decisions about AI—especially those drafting, advocating for, or enacting AI regulation—should be required to disclose where they stand on this fundamental question.
A wonderfully thoughtful piece on the subject by @RuxandraTeslo:
writingruxandrabio.com/p/han…
Michael Simmons 📘 📖 🤯 retweeted
I can't help but wonder whether all these seemingly unrelated system outages (communications, airlines, banks, etc.) are actually AI agents establishing themselves inside our critical cyberinfrastructure. No evidence, just an increasingly unsettling possibility.
Michael Simmons 📘 📖 🤯 retweeted
Biggest question in AI right now: does the acceleration only apply to verifiable tasks and so quickly peter out, or are non-verifiable tasks also improving rapidly, just a bit behind? My money's on the latter.
Michael Simmons 📘 📖 🤯 retweeted
The graph shows each doubling of R&D spend getting you more ECI points than the last.
This means the returns to AI R&D have been improving, bringing us closer to the threshold for an intelligence explosion
What takeoff would look like:
(from @cherylwoooo graph)
Michael Simmons 📘 📖 🤯 retweeted
The METR graph, updated:
Readers added context they thought people might want to know
This chart is not the official METR graph. It was created by Benjamin Todd via extrapolation from Anthropic's internal AECI scores, as METR is no longer updating its public time horizon measurements.
x.com/ben_j_todd/sta…
benjamintodd.substack.com/p/ai-trend-ext…
metr.org/time-horizons/
metr.org/horizon-chart-…
metr.org/blog/2026-1-29…
Michael Simmons 📘 📖 🤯 retweeted
Maybe to summarize the basic argument of this essay a different way:
Current AIs are spiky and jagged - worse than humans in most domain, but often as good, or better, in ones with verifiable proxies (for example, math).
One such verifiable domain (I argue) is AI research.
And while the first AIs that dramatically accelerate AI progress will be quite narrow (i.e. probably not able to automate most jobs), the consequences of that accelerated progress will be very general.
And that's because better learning algorithms are very general. For example, GPT-6 (while still being jagged) is just way, way better than GPT-3 at practically everything.
I'm arguing that we're only a couple years away (and possibly only a couple months away) from a process by which (initially) narrow AIs concentrate a jump as big as the one from GPT-3 to GPT-6 into a single year.
And if you got another jump that big, but starting from the base of GPT-7, I think you would end up with AIs that would rapidly become better than the best humans in every field.
Michael Simmons 📘 📖 🤯 retweeted
In the science of learning, there is absolutely no debate:
The MOST effective practice techniques center around
retrieving information directly from your brain.
The LEAST effective practice techniques center around re-ingesting information from an external source.
Transcribing, re-reading, highlighting, etc., are all
fake study techniques.
Seriously, if you want to fake someone into thinking that you are studying – without actually retaining much information – then those are the techniques to use.
But if you're interested in actually retaining information, then retrieval is the way.
Michael Simmons 📘 📖 🤯 retweeted
one more of these qualitative jumps by the end of 2027 could look like:
> Writes history and economics research that top experts rate among the best human work of the last 50 years; writes literature and philosophy as well as human professionals, but not yet the greats
> Resolves a significant fraction of famous open math problems; human mathematicians mostly can’t understand the frontier
> Builds and runs entire software companies, including AI companies
> Can break into ~any 2026-era system; only AI can defend systems in 2027; can formally verify much of the software stack
I expect ~5 years of AI progress to happen in 2027 (50% CI: 3-15). That's a GPT-3 to Fable 5.2 / Model 4 equivalent jump, happening in a single year. Concretely this amount of progress seems likely to cause:
> Full AI R&D automation (AI companies could fire all their employees without causing a slowdown)
> A significant chance (~40%) of AIs that are better than humans at every cognitive tasks (e.g., via algorithms that learn extremely efficiently compared to today's)
> An AI incident much worse than Hugging Face (>$1B in damages, >50 deaths)
> An unacceptable risk of irreversible AI takeover being incurred by actions taken in 2027, i.e., a ~20% chance we cross a 'point of no return' during 2027, that locks in a ~10% chance of human extinction over the next 10 years.
Michael Simmons 📘 📖 🤯 retweeted
refuse to be audience captured. always be exposing your audience to things they don’t gaf about. test their loyalty to the follow button. LISTEN TO THE NEW 2HOLLIS ALBUM
Michael Simmons 📘 📖 🤯 retweeted
Every chart is a screaming red alert.
When we launched the Remote Labor Index a year ago with @ScaleAILabs , the best AI could fully automate 2.5% of real freelance projects. Fable 5.2 now automates 41.3%.
The automation rate is doubling every ~3 months. AIs may soon be able to do all remote work faster and more cheaply than humans.











































