Umair S

@satnam6502 ∀ context ∈ {winter, summer, breakfast, dinner, sadness, celebration}, ghee_on(makki_roti) > ghee_on(wheat_roti) Proof is easier than we think so QED

An enduring childhood memory is of my mother making a vast vat of ghee, slowly melting 35lbs of butter and skimming off the impurities to yield liquid gold. It filled our apartment with an overwhelmingly comforting nutty sweet aroma, and even today the smell of melting butter transports me to that small flat in Glasgow and a feeling of security and love.

@ShriramKMurthi This recent paper from Anthropic might be relevant to the topic of your talk, in case you haven’t seen it yet: anthropic.com/research/AI-as…

New talk abstract dropping. I just hope I can write a talk to live up to it by *checks watch* *gulp* Monday.

I recently asked the Harmonic API to prove Z-Transform theorems (en.wikipedia.org/wiki/Z-transfo…). It performed really well. This is exactly the work from my ITP-14 paper using HOL Light. By having Z-Transform formalization and proof of its properties, we can reason about quite a few concepts from digital signal processing and control systems. Soon I will share my experiments. @HarmonicMath

A few months ago when I started to learn Lean I made the decision to work my way through the "Functional Programming in Lean" tutorial but to totally ignore the "Theorem Proving in Lean" tutorial. That's because all I want to do is to be able to state theorems. I'm leaving the proofs to @HarmonicMath's Aristotle system. It's amazing how well this mad strategy is working out in practice. Theorems For Free! #Lean4

@DrJimFan Right, but in the context of Physical AI, two important open questions are the latency and accuracy (which are somewhat relaxed for LLM applications).

@ShriramKMurthi @willowtv I gave up and got a new subscription to watch the game.

Hey @willowtv — your new software platform has made login fail everywhere. Can't log into Willow on my iPad NOR on my Roku. The Roku version now *forces* me to subscribe through Roku, even though I already have a sub. Seriously buggy — better rollout please?

When AI can generate a thousand working implementations before lunch, the question shifts from “can we build it?” to “did we ask for the right thing?” Upper-level PL courses become training grounds for precision specification - teaching students to write specs so airtight that AI can’t wriggle through loopholes with technically-correct-but-useless solutions.

In a world where most code in modern programming languages will be machine-generated, what is the role of an upper-level programming languages course? Interesting and non-obvious answers please.

Yesterday, I sat down with my daughter (grade 1) to brainstorm an app idea. Instead of jumping straight to features, she started with something simple but profound: "First, let's talk about what we DON'T want." - No games - No work apps - No boring apps Then she moved to what we DO want: a voice-driven teaching app with folders for different subjects: math, reading, art, music. She even mapped out the sign-in flow (it's funny that she added brother/sister name and pet name 😃), thinking through how kids would actually use it. She didn't need to know how to code. She needed to know how to think. This is the power of first principles. LLMs are useful tools. But they're most effective when we bring: - Clear problem definition - Strong conceptual thinking - A well-structured foundation The future belongs to clear thinkers who know how to ask the right questions.

One downside of focusing AI safety discussions primarily on extreme scenarios is that it can detract from examining AI safety through the lens of system safety and technological risk management. It's unfortunate that advertisement of certain terminology in AI communities has led to overly narrow thinking about safety concerns.

@JundeMorsenWu @ziv_ravid this is related to the concept of “Free Riders” 😀

Let's talk a bit about authorship order on a paper. Yes, everyone cares about it, and it can become very emotional. Even if you think that big professors don't care, they do (although I know two professors who don't—you can probably guess who).

We’ve found as AIs get smarter, they develop their own coherent value systems. For example they value lives in Pakistan > India > China > US These are not just random biases, but internally consistent values that shape their behavior, with many implications for AI alignment. 🧵

Can we trust AI? “Everyone in the world is getting onto this brand new airplane that really has never been tested before. And it’s going to take off and it’s never going to land. It has to fly, forever.” – Stuart Russell, IASEAI President.

Almost everyone I speak with wants tool AI rather than AGI: controllable AI than empowers us rather than overmarketed “digital god” AI that replaces us and that we don’t know how to control. The key point about the Venn diagram below is they we can get basically all the tools we want as long as we don’t put too much “A”, “G” and “I” into the same system:

Fully agreed, most of the benefits that we want are in A, G or I.

As AI models rapidly advance in capabilities, their impact on people's lives will only continue to grow. It's crucial that governments prioritize the well-being of citizens and ensure they listen to them as they shape the future of AI. time.com/7213096/uk-pub…

Concluding #IASEAI25, we share our thoughts on critical imperatives discussed, ahead of AI Action Summit. These items aim to provide standards, regulation, international cooperation, and research funding, ensuring the AI lifecycle is safe, ethical, and beneficial to all humanity. #EthicalAI #AIRegulation #TechPolicy #AISafety #SustainableAI #AICooperation

Thank you to all our speakers, presenters, and attendees at #IASEAI25! We look forward to continuing the dialogue we started this week. We’ll post our recorded sessions in the coming days. Join the movement to ensure AI benefits all of humanity. #AISafety #EthicalAI #AIRegulation #TechPolicy #SustainableAI #AICooperation Special thanks to our plenary speakers: @MathiasCormann , @Yoshua_Bengio , @ancadianadragan , @geoffreyhinton , @mariaressa , @mmitchell_ai , @JosephEStiglitz , @tegmark , @katecrawford , @Dr_Atoosa , @ghadfield , @tobyordoxford , @NicolasMoes , @zicokolter

Our CEO, Umair Siddique @imumairs , will represent @reasonxlabs at the @IASEAIorg '25 Conference. He’ll join global leaders, incl. Nobel laureates & top researchers, to discuss AI safety & ethics. His talk: Safety of AI-Powered Robots: Challenges & a Way Forward.

@satnam6502 dspotdessert.com have some good options (at least in Ottawa)

I am spending a few days in Toronto at the Groq office. I’ve already identified Almond Butterfly Cafe as a great looking gluten free celiac option. Looking forward to discovering more.

@jhasomesh may be he wanted to say \ln(100) 😂

A department chair told me that they hired someone who already has more than 100+ papers. What! How is this possible? Thoughts.

@TzuHanH Congratulations 👏

‼️New Papers‼️ Very excited to share that we got 2 ✌🏻 papers accepted to TACAS’23!!!🫣🥹😍🥳 Big thank you to all my collaborators. Submitted 2 and get both accepted feels truly magical🪄(titles in the reply.) Going to be in Paris in April the very first time in my life!🤩