←
Alex Nemish 🇺🇦1,479 posts
Alex Nemish 🇺🇦 avatar

Alex Nemish 🇺🇦

@atlanter

Creator of @Scalus3. Founder @lantr_io. Cardano DRep. Ex-IOG. Wingsuit formation world record holder.

◷ Joined February 2009▧ 181 Photos and videos
Alex Nemish 🇺🇦 avatar
Alex Nemish 🇺🇦@atlanter·Aug 15

In June, 26 of us from 9 countries set 5 FAI records at Skydive Balaton, Hungary, including a 25-way sequential wingsuit formation world record. 🇵🇱🇩🇪🇱🇻🇮🇹🇺🇦🇫🇷🇨🇭🇪🇸🇸🇪 The plan was one European record. We set five. The story started with lost luggage.

0:45
Serhii Sternenko avatar
⟳ Alex Nemish 🇺🇦 retweeted
Serhii Sternenko@sternenko·Nov 17

Ukraine destroyed Crimean bridge. Just help Ukraine defeat Russia. We can do it.

Post attachment
Steeve Morin avatar
⟳ Alex Nemish 🇺🇦 retweeted
Steeve Morin@steeve·Nov 9

the Jensen scandal wouldn't work in France

0:49

BREAKING: Nvidia, $NVDA, CEO Jensen Huang sends a company-wide email regarding his recently reported relationship with a Chinese spy.

Post attachment
Alex Nemish 🇺🇦 avatar
Alex Nemish 🇺🇦@atlanter·Nov 4

ZK proofs of LLM inference on a 400B model. Really cool. Now imagine this on LeCun’s far smaller JEPA models: verifiable inference at a reasonable cost.

New Anthropic research: progress on provable inference for large language models. Here, we introduce FlashProve. Building on established cryptographic techniques, FlashProve allows developers to verify whether a given response came from a given AI model. It’s substantially more efficient than prior open-source methods for doing so. These proofs have a range of applications. They can, for instance, make it harder for attackers to hijack a chat, or enable safety evaluators to verify that the model they’re testing is the same one being served to customers. For our full write-up, see here: anthropic.com/research/prova… Further progress in provable inference will likely be an important part of improving trust in AI deployment. To that end, we're open-sourcing FlashProve’s code, here: github.com/anthropics/FlashP… We’re grateful to our collaborators: Alessandro Chiesa, @GiacomoFenzi, @gwberg95, @miranda__christ, Rahul Dalal (rdalal.github.io) and Ben Mayer (@transmissions11).

Alex Nemish 🇺🇦 avatar
Alex Nemish 🇺🇦@atlanter·Nov 4

At Scala Days, Martin told me about a prototype of the Scala 3 compiler in WebAssembly. We found it, added Scalus, and built play.scalus.org – write Cardano smart contracts in Scala 3, compiled entirely in your browser. Highly experimental, but it works!

Alex Nemish 🇺🇦 avatar
Alex Nemish 🇺🇦@atlanter·Nov 3

Could someone please reverse-engineer MacOS and reimplement it open source on top of Linux kernel? It seems possible this days 😂

In a perfect world Apple would use Linux kernel and open source/sell MacOs independently from its hardware. That would vanish Windows existence in 5 years and make Linux the single OS for mobiles, desktops, servers, and supercomputers. #Apple, please! #Linux #PerfectWorld

Yann LeCun avatar
⟳ Alex Nemish 🇺🇦 retweeted
Yann LeCun@ylecun·Nov 1

LLMs accumulate a lot of declarative knowledge. That's how they can answer a large number of questions. That's also why they require so many parameters and hence so much memory. But one can replace the accumulation of knowledge by *understanding* That what JEPA world models do. That's one reason they can be useful with considerably less memory and training data than LLMs. This paper merely shows that the underlying similarity structure of the data, whether text or image (e.g. co-occurence of words in text or objects in images) uniquely determines the shape of the representation point cloud, when the encoder is trained trained with a common anti-collapse criterion like SIGReg or VCReg (not by reconstruction).

Rikiya Takehi avatar
⟳ Alex Nemish 🇺🇦 retweeted
Rikiya Takehi@rikiyatakehi·Nov 2

We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…

Post attachment
Alex Nemish 🇺🇦 avatar
Alex Nemish 🇺🇦@atlanter·Oct 30

I presented this approach last year in Edinburgh, at UPLC 2025, IOG's conference on Plutus Core and Cardano's programming languages: iog.io/news/uplc-2025-confer…

Alex Nemish 🇺🇦 avatar
Alex Nemish 🇺🇦@atlanter·Oct 30

The code: • tiny-uplc-jit: the three engines from the talk, in a few hundred lines github.com/lantr-io/tiny-upl… • the production staged JIT in Scalus, with exact Plutus budgets github.com/scalus3/scalus/tr…

Alex Nemish 🇺🇦 avatar
Alex Nemish 🇺🇦@atlanter·Oct 30

The article shows the whole approach in steps: a CEK interpreter, a closure compiler (16×), then Scala 3 quotes and splices (35×), with all the code, the generated Scala, and two sets of benchmark numbers. lantr.io/blog/blazing-fast-l…

Alex Nemish 🇺🇦 avatar
Alex Nemish 🇺🇦@atlanter·Oct 30

Blazing-Fast Lambda Calculus via Metaprogramming: my ScalaDays 2026 talk is now an article. Scala 3's staging compiler turns a lambda-calculus interpreter into a JIT in a few dozen lines. In @Scalus3, it runs Cardano smart contracts ~3× faster than the Haskell VM 🧵

Post attachment