ashish

writing about coordination in supercritical systems, document intelligence @datalabto, helping launch collective flourishing @aria_research, previously @scite

[Evan Reiser @evanreiser]

.@AnthropicAI has filed a lawsuit against @Abnormal
claiming we copied their brand to mislead security customers, and they are asking for “all revenues, earnings and profits”

This is obviously not true. They don’t own every A/slash design in AI cybersecurity, and they don’t get to turn a logo dispute into a claim on Abnormal’s entire business.

It would be easier to concede quietly. But Abnormal was built on trust, innovation, and intellectual honesty.

Values are what you do when tested (especially when inconvenient) Our behavioral security platform and our AI models are ours.

Everything meaningful about Abnormal was built the hard way: the technology, the customer relationships, and the trust. We earned that trust by protecting customers, and we will defend it.

My blog post about it (link below):

[ashish @supershishy]

@paulg white vintage dials are also nice because they all age in fun ways -- champagne, off white, some of mine turned pink. Way more charming than black ones in person.

[ashish @supershishy]

@sebkrier The best alpha is pre-narrative and will come from familiarity and experience grounded in substance in a domain!

[ashish @supershishy]

@jasoncrawford i built my own note app that syncs across devices and has background agents suggesting ideas and reading (it's all attributable, so I can check see what it added, but also ignore it easily). can group notes to enter a research / writing workflow w agents or myself.

[Jordan Dworkin @jddwor]

If half of scientific fields are accelerated 5x by AI and half aren’t, is that better or worse than if all fields are accelerated 3x? And how should you allocate resources in that scenario to maximize discovery? Proportionally? More to the accelerated fields given their potential for rapid progress?

In a new piece for the Abundance and Growth Blog, I look at what happens to scientific progress when AI speeds up some fields but not others (link in the next tweet).

Jagged capabilities are familiar from the task and productivity literature, and a series of recent papers demonstrate that when automatability varies across tasks, overall productivity gains are constrained by the weakest ones.

This applies to scientific and research tasks as well, as @bfjo shows in a recent working paper. But science is likely to face an additional layer of this dynamic one level up:

Discoveries are not only bundles of tasks, but also inputs to other discoveries. And scientific progress depends on aggregation and cross-pollination of discoveries across disparate fields. So if AI accelerates some fields much more than others, then that interdependence may turn less-accelerated disciplines into long-run bottlenecks for accelerated ones.

In the broader economy, this kind of aggregation bottleneck (e.g. in supply chains) is often made visible by price increases and resulting investment in scarce inputs. But science lacks clean, contemporaneous price signals (especially prospectively) for how interrelationships between fields are driving or blocking progress. What is visible – e.g. novelty, prestige, the recent pace of progress in a field – could even lead to investment shifting away from the pace-setting weak links rather than towards them.

To make these ideas more concrete, and to allow you to play with the underlying structures and assumptions, I built a simple model of discovery as a network of interdependent fields, accelerated unevenly by AI (dashboard link in the next tweet).

Across a range of assumptions, the results show that uneven acceleration across fields reduces overall progress, relative to a comparable uniform case, by over 50%; they also show that disproportionate investment in the accelerated fields can worsen this bottleneck, while modest overinvestment in unaccelerated fields can partially alleviate it.

Among other things (more in the piece), I think this dynamic makes a case for investing in AI-driven acceleration where possible, while retaining broad and deep support across science even when the short term returns are not obvious and the long term returns are uncertain.

[Schmidt Sciences @schmidtsciences]

How do we ensure safety in a world with millions of interacting agents, built and deployed by many different actors?

Learn more about our new research fund in partnership with @GoogleDeepMind @Googleorg @coop_ai @ARIA_research:

https://t.co/UmbGd2wev0

[KnicksMuse @KnicksMuse]

This is the clip.

[ashish @supershishy]

@growing_daniel it's not over until its over!!

[Fei-Fei Li @drfeifei]

Scientific research is fundamental to advancing civilization and helping people globally to solve the most critical problems, from medicine to materials, from brain science to physics, and much beyond. This is only possible when scientists have access to the best tools of the time to conduct scientific research, including having access to AI-based tools.

[Markus J. Buehler @ProfBuehlerMIT]

We've made a breakthrough in self-evolving AI scientists moving from "search" to "principled discovery": Scientific discovery requires that the search space itself changes, and an AI scientist must perceive this shift without intervention. We built an AI that achieves this for the first time with the ability to discover the scientific vocabulary it reasons in. Evidence, tools, artifacts, verifiers, failures & claims become typed provenance. We show three distinct modalities: 1) retrieval, adding known objects; 2) search, exploring a fixed schema; and critically: 3) discovery, a verified regime transition.

We solve the open-endedness evaluation problem by lifting agentic workflows into a typed copresheaf and proving, via a Kan obstruction, that true discovery is not unbounded generation but a verifiable schema expansion: old evidence is transported by Left Kan extension, and genuine novelty is mathematically quantified by the pointwise residual beyond the transported image - separating discovery from mere search and making novelty objective and measurable rather than a subjective judgment or benchmark delta.

Our AI scientist is built in a way that does not pre-conceive the approach it chooses; instead, we endow the system with formal power to adapt, evolve, and reason from first principles. Case studies include:
1⃣Builder/Breaker model that discovers mode-conditioned compliance in proteins;
2⃣CategoryScienceClaw that finds anisotropic fiber-network stiffness rules.

Great work in collaboration with my graduate student @fwang108_ @MITdeptofBE

F.Y. Wang & M.J. Buehler, Self-Revising Discovery Systems for Science: A Categorical Framework for Agentic Artificial Intelligence, arXiv:2606.01444, 2026

[ashish @supershishy]

@sebkrier submitting mandated siestas as a policy intervention

[Fei-Fei Li @drfeifei]

https://t.co/Kt50ttQRMJ

[ashish @supershishy]

@James_West_PhD @akoustov @BrankoMilan Aye aye, there's some more recent thinking from publishers on this, but it's imo not forward looking enough. It's also yet another coordination problem 😊.

[ashish @supershishy]

@James_West_PhD @akoustov @BrankoMilan Yeah -- I think theres a gap in tooling and infra to streamline that. It should honestly just happen without extra work for a researcher, but we're a bit far from that today (depending on one's discipline)

[ashish @supershishy]

@James_West_PhD @akoustov @BrankoMilan one thought is to stop replying on the figure itself and enforce provenance and attestation for the full chain of custody. a supply chain perspective is helpful here (see: https://t.co/AEl3ydfANE)

[ashish @supershishy]

@theandrewsiah i love that chicken rice shop

[Séb Krier @sebkrier]

@1thousandfaces_

[ashish @supershishy]

@stuartbuck1 that IPOs and public markets are dumping grounds

[maja 🔭🍒 @majamediaco]

writing is one of the oldest technologies for altering consciousness

[Edge City @JoinEdgeCity]

The full menu of experiments at @EdgeEsmeralda 2026 🧬

Free mass spec, a month of wearables, a jhana retreat, a vasocomputation study, a 4-week body composition model, an ADHD morning routine prototype, and the Alethios platform for your own studies.

Sign up below ↓