←
Basis102 posts
Basis avatar

Basis

@BasisOrg

Basis

◷ Joined July 2022▧ 7 Photos and videos
Basis avatar
Basis@BasisOrg·Jul 7

We're at ICML this week presenting work on world models, dynamical systems modeling, and more. If you're here and would like to discuss over a drink, come to our social on Thursday evening! Link below.

Post attachment
Basis avatar
Basis@BasisOrg·19h

📣 Applications are open for our Summer 2027 research internship program at Basis. We’re looking for graduate-level researchers and candidates with equivalent experience to work alongside our scientists and engineers. Interns will develop computational theories of intelligence, distill insights from solving problems into more general mathematical and computational theories, and contribute to open-source software. Depending on the project, you could: • Develop methods for learning the dynamics of complex systems from data and quantifying uncertainty • Develop formal methods to verify that scientific software continues to behave as intended as it evolves • Model how social animals coordinate in the wild • Develop sensing systems that preserve privacy in public spaces • Build robots that learn by running experiments Internships are full-time and in person in New York City or Cambridge, MA. Start dates are flexible, and internships last at least three months. Applications will stay open until the cohort is full. Learn more and apply here: jobs.ashbyhq.com/basis-resea…

Emily Mackevicius avatar
⟳ Basis retweeted
Emily Mackevicius@e_mackevicius·19h

Thanks for highlighting our work! @foresightinst Vision weekend was magical, I really appreciate you pointing out the strong female presence at this event imagining and co-creating hopeful futures.

Vision Weekend is by far the most woman-forward frontier tech event I've ever been to. Nobody's said it out loud yet, so I will. @foresightinst was co-founded 40 years ago by Christine Peterson (@lifeext). Yes SHE came up with the phrase "open source" we still fight about what that means in AI! It's run today by a women-led team, CEO @allisondman, COO @KarolinaEklow, event lead for Vision Weekend Lydia La Roux, plus Tara Padovan-Hickman and Program Director @madvinuova. Together they made a 500 person event spanning two venues look effortless. Allison has built a culture where women at the top of science are the norm. 15 of the 34 speakers on the main program were women and these weren’t junior slots. Many were PhDs running their own labs and missions across longevity, neurotech, AI, space and molecular machines and they brought some pretty big ideas… - @ltelkins, the scientist behind NASA's Psyche mission, talked about shielding from cosmic rays - including the Oh-My-God particle - a single particle detected in 1991 with MILLIONS of times more energy than anything we can produce in particle accelerators on Earth. - @erika_alden_d believes terraforming Mars is possible in our lifetime - Pioneer Labs has already produced the first microbe able to convert Martian regolith, water and air into bioplastics for potential habitat construction. - @e_mackevicius, co-founder of @BasisOrg, a non-profit working to develop AI as public-interest infrastructure, also researches rat collective behaviour and what it can teach us about agent swarms. - @zweinberg of @buildexante is investing in human agency and backing startups working on privacy, data ownership, security and identity. Tomorrow at Lighthaven we'll be recording some podcasts with @naomibrockwell and @NicoleOzer, two of the leading voices on privacy and digital rights. I'm immersed most days in Local AI, open source AI and open-weight models and our spaces tend to look a bit similar (I'm as guilty as anyone). There’s no posturing at Vision Weekend. People engage in meaningful discussion about the most important topics. Spaces Ike these provide role models and experiences to people who wouldn’t normally think of themselves as potential leaders in frontier tech. If we're serious about AI for everyone then maybe our spaces should start to look a little more like that. Thank you @foresightinst. One more day to go and I'm already more optimistic than when I arrived!

Post attachment
Basis avatar
Basis@BasisOrg·20h

We built an undo button for Amazon S3. Blog: basis.ai/blog/undo-for-agent… Before letting an AI agent make a change, our prototype records how to reverse it. If it can’t undo the operation, it blocks it. Try it in our interactive demo. Change the counter, then click the Rollback button to return it to its starting value. Thanks to the AWS Automated Reasoning Group for supporting this research.

With support from AWS and the Automated Reasoning Group, at @BasisOrg we've completed a research prototype of infrastructure for provably safe agent actions: an undo button for S3. Blog: basis.ai/blog/undo-for-agent… We run agents in a setting where, if an agent makes a mistake, we can roll the system back to a previous state. This is hard because S3 is extremely complex, with many layers of data, permissions and configuration. To do this we needed two things. First, we synthesized a "compensator" for the S3 API. A compensator is a program that intercepts calls from the agent to S3 and stores the state needed to undo each operation. To synthesize compensators we put LLMs in a game: one writes tests that break the current compensator, and another repairs it. Second, for our compensators to carry formal guarantees, we need to verify them against a model. We built a partial model of S3, and added a verification step to the synthesis loop. This is a research prototype. With further work it could be put into production to bring more safety to the billions of API calls that go to cloud infrastructure everyday. Even more importantly, the approach extends far beyond S3 and undo. We think it's possible to realise a world where most of the world's infrastructure is continuously, formally modeled, and agents prove in real-time that their actions are safe from many kinds of harm. Research done by Jack Feser and @tweeting_nikhil

0:43
Zenna Tavares avatar
⟳ Basis retweeted
Zenna Tavares@ZennaTavares·22h

With support from AWS and the Automated Reasoning Group, at @BasisOrg we've completed a research prototype of infrastructure for provably safe agent actions: an undo button for S3. Blog: basis.ai/blog/undo-for-agent… We run agents in a setting where, if an agent makes a mistake, we can roll the system back to a previous state. This is hard because S3 is extremely complex, with many layers of data, permissions and configuration. To do this we needed two things. First, we synthesized a "compensator" for the S3 API. A compensator is a program that intercepts calls from the agent to S3 and stores the state needed to undo each operation. To synthesize compensators we put LLMs in a game: one writes tests that break the current compensator, and another repairs it. Second, for our compensators to carry formal guarantees, we need to verify them against a model. We built a partial model of S3, and added a verification step to the synthesis loop. This is a research prototype. With further work it could be put into production to bring more safety to the billions of API calls that go to cloud infrastructure everyday. Even more importantly, the approach extends far beyond S3 and undo. We think it's possible to realise a world where most of the world's infrastructure is continuously, formally modeled, and agents prove in real-time that their actions are safe from many kinds of harm. Research done by Jack Feser and @tweeting_nikhil

0:43
Zenna Tavares avatar
⟳ Basis retweeted
Zenna Tavares@ZennaTavares·Nov 11

For the past year @BasisOrg has been building a general framework for probabilistic reasoning in dynamical systems, called Dynestyx. Blog: basis.ai/blog/dynestyx/ Dynamical systems are the workhorse of science and engineering: robotics, epidemiology, climate, biology, etc. The basic thing you can do with a dynamical system is simulate it: run it forward to compute how it will evolve over time. Dynestyx goes beyond this and lets you do inference and reasoning in dynamical systems. This includes: - Forecasting: What will happen next, and how sure are we? - State inference: Given noisy, partial measurements, where has the system actually been? - System identification: How does the system work? What parameters and rules produced the data we see? - Causal reasoning: Given the evidence, what would happen if we intervened on the system? Dynestyx allows you to do probabilistic inference over all kinds of dynamical systems (ODEs, SDEs, PDEs). You (or an AI agent) write the model once, then swap in different inference methods with minimal changes. Its generality allows us to tackle many different systems, as shown in the GIFS: - recovering a high-res Navier–Stokes fluid flow from noisy low-res measurements - steering particles through the wake behind a cylinder - a robot hand turning a cube it can only see through a camera - forecasting a 1978 flu outbreak at a boarding school as the data comes in Also: it's open source. Please give it a try: github.com/BasisResearch/dyn… Led by Matt Levine and @DanWaxmanML, with many contributions from our research staff

Basis avatar
Basis@BasisOrg·Nov 10

When learning dynamics from data, changing your models or inference methods shouldn’t mean rebuilding your analysis. Meet Dynestyx, our probabilistic programming library for dynamical systems. Our latest blog explores why we built it and how it helps researchers access an unprecedented suite of state-of-the-art strategies for learning dynamics with quantified uncertainty. Read more: basis.ai/blog/dynestyx/

Basis avatar
Basis@BasisOrg·Nov 6

What does the world sound like to a rat? 🐀 This Halloween, attendees of the City Creatures event in SF donned rat-ear headphones to hear ultrasonic calls from Basis’ urban rat research, made audible to humans. Learn more: basis.ai/news/hidden-sounds-…

Post attachment
Max von Hippel avatar
⟳ Basis retweeted
Max von Hippel@maxvonhippel·Oct 26

I am working on a map of the secure program synthesis space, in collaboration with @sergeybratus for our NDIST keynote on Tuesday. Here's a sneak-peak. The words "RISC-V" and "Lean" are used here WLOG, and some simplifications are made to fit everything on one page. Did I miss anything? 1. Develop semantics for each language in an interactive theorem prover, and then use these semantics to prove that every program is correct. (@oath_tech has done some work along these lines. See in particular Sam Xifaras' project on Ruby samx.io/blog/topics/devlog/2…) 2. Develop semantics for an architecture, say RISC-V, or perhaps for all possible architectures, and then use this to prove that every compiled program is correct. (@AnthropicAI made headlines recently with their announcement of a very large-scale project along these lines in collaboration with @theoremlabs anthropic.com/research/forma…) 3. Develop semantics for an architecture, say RISC-V, or perhaps for all possible architectures, and then use this to capture a formal model of each existing compiler or interpreter, and then iteratively (provably) refine each such model to an interpretable and easy-to-use formal semantics; then continue as in approach 1. (See foundational work by @kirancodes at BASIS AI basis.ai/blog/ship-your-inte…) 4. Make Lean, or some other such ITP + programming language, fast to run (comparable to Rust or C), and then completely rewrite all software in it, and prove the rewrites are correct. (This is I think, loosely speaking, the idea behind @VictorTaelin 's Bend.) 5. Make implementation-level verification tools, such as Verus, more capable and ergonomic, and then use them to prove all software correct. (This is what @fhuszar and @Craig_Hughes_ are doing at Reasonable, & I'm aware that BASIS and AWS are also working on this.) 6. Design a language in which unsafe states are illegal, and then rewrite everything in that language. Here, what kinds of safety you can express will be much more limited but the problem is, conversely, way more tractable. (This is what @danwallach and @DARPA did with TRACTOR, where the safety property is memory safety.)

Post attachment
Basis avatar
Basis@BasisOrg·Oct 26

Heading to DCASE2026 this week? Catch our co-founder @e_mackevicius's keynote, “Who Is the Listener? Privacy-Aware, Behavior-Grounded Bioacoustics in Urban Rats,” at 9:30 a.m. Thursday! dcase.community/workshop2026…

🐀🎙️Really excited to give a keynote at @DCASE_Challenge workshop next week 🎙️🐀 Talk title: Who Is the Listener? Privacy-Aware, Behavior-Grounded Bioacoustics in Urban Rats @NJourjine is also presenting a related poster at BioDCASE tomorrow! biodcase.github.io/workshop2…

Basis avatar
Basis@BasisOrg·Oct 23

Basis researcher @ralph_ep spoke at PestWorld 2026 this week about how AI and field neuroscience can help us understand urban rat behavior and inform rodent monitoring and mitigation. Thank you to the pest management community for the warm welcome and thoughtful conversations about this work! Photo credit: Bobbie Wilson, Pestec

Post attachment
Basis avatar
Basis@BasisOrg·Oct 20

Basis researcher @ralph_ep will be in Texas this week to give a talk at PestWorld 2026! 🐀 Catch his talk, "Eavesdropping on Urban Rats: Bringing AI and Field Neuroscience to Rodent Monitoring and Mitigation," on Thursday 10/22 from 1:30-2:45 PM in Fort Worth 6-7.

Basis avatar
Basis@BasisOrg·Oct 20

Most programming languages lack the precise rules software verification requires—a growing bottleneck as AI gets better at writing proofs. @kirancodes used LLMs to write those rules for a small language and proved in Lean that they match how the code runs. basis.ai/blog/ship-your-inte…

For a while now, @kirancodes has been excitedly telling everyone at @BasisOrg who would listen about a crazy idea she's been pursuing. Not everyone thought it was possible, but it's starting to work. See basis.ai/blog/ship-your-inte… TL;DR: The bottleneck in verifying the world's software is shifting from theorem proving to specifications and semantics. We now have a way to verify semantics that can scale across all programming languages. The basic idea: To verify that a program is correct, you need the program, a specification you want to prove, and a semantics that defines what programs and specs mean. Most programming languages (Python, Ruby, Lua…) have no formal semantics, because semantics are painful to write for complex languages. You can ask an LLM to write a semantics for you, but it'll probably have mistakes. But (i) programming languages do have interpreters/compilers, which ultimately are compiled down to machine code, and (ii) some instruction sets do have formal semantics. RISC-V has an official formal model. So, crazy idea: Step 1: Compile your interpreter (e.g. Python's) down to RISC-V. Now its behaviour is formally defined. It's exact, but unreadable: thousands of machine instructions. Step 2: Ask your LLM to write a semantics for your language. Step 3: Ask the LLM to prove, in @leanprover, that its semantics agrees with the compiled interpreter on every possible program. If the semantics is wrong, the proof gets stuck, and the LLM goes back to fix it. Step 4: If the proof goes through, you have a semantics you don't need to trust, and you can verify programs in a language that never had one, with guarantees that reach the binary that actually runs. Kiran has now done this end to end for a simple WHILE language, and we're scaling it up to Lua and OCaml, with Python as the north star.