←
Shift17,949 posts
Shift avatar

Shift

@Shiftreduce
◷ Joined June 2011▧ 297 Photos and videos
John LeFevre avatar
⟳ Shift retweeted
John LeFevre@JohnLeFevre·Nov 16

The SpaceX engineers who caught the rocket don't write LinkedIn thought pieces about leadership.

Charmi avatar
⟳ Shift retweeted
Charmi@Charmi·Nov 10

NO PUEDE SER, ESTO ES UNA AUTÉNTICA BARBARIDAD 😭💀 Han conseguido transformar POKÉMON ROJO FUEGO EN UN JUEGO COMPLETAMENTE EN 3D. Han recompilado el código original de GBA, lo han reescrito en RUST y lo han combinado con los gráficos y animaciones de Pokémon Let's Go. ¿LO MÁS LOCO? Los combates, la IA y las interacciones siguen funcionando con el código original de Rojo Fuego, pero ahora tenemos CÁMARA LIBRE Y UN MUNDO 3D AMPLIADO. ESTO ES LITERALMENTE EL REMAKE QUE TODOS QUEREMOS 😭

Post attachment
Gal Beniamini avatar
⟳ Shift retweeted
Gal Beniamini@laginimaineb·Nov 6

What comes after the transformer? Over the last couple of months, we've been running these crazy interesting experiments, letting LLMs redesign neural nets as they train. The architectures that emerge are an incredibly strange mixture of learned weights and spliced code. I really recommend reading this one carefully. Very proud to have been part of this research.

1/ What if an LLM could not just edit a model's weights, but re-architect it as it trains? We tried just that. The resulting architectures exit the known territory of transformers: the coding agents spliced code directly into the network. On nanochat, the loss fell below ordinary training after each new code layer was introduced. On algorithmic reasoning, accuracy rose from 55% to 84%. We even let DeepSeek-V4.1-Flash act as the agent modifying itself, raising its AIME and HMMT 2026 score from 64% to 71%. To do all this, the agents built a set of tools we’ve never seen before, which we call mechanistic intervention! Blog: doubleai.s.gy/llm-redesign

Post attachment
Sean Heelan avatar
⟳ Shift retweeted
Sean Heelan@seanhn·Oct 30

haha, so apparently OpenAI are about to start training on your Daybreak Blue sessions, with no way to opt out. Literally stealing people's private work, as they do it, is cartoonishly evil. It also will presumably kill Daybreak usage by security researchers. Or at least, it should. The thing OAI have had to swear they are not doing to math researchers, they are now explicitly about to start doing to security researchers. Yet another reminder of the importance of open models.

blasty avatar
⟳ Shift retweeted
blasty@bl4sty·Oct 28

btw now would be a good time to build the new hackcoza/milw0rm/exploit-db that doesn't suck and is correctly maintained/curated. throw a shitton of clank and compute at it for e2e verification of all hax. community notes/dispute system for QA etc. etc.

Benjamin F Spector avatar
⟳ Shift retweeted
Benjamin F Spector@spectorb·Oct 21

Hello again, it's been a while. We're writing to let you know that we have raised quite a lot more money. We're letting you know because we are on the lookout to hire smart people and this will help us do that. See you again when we have a product or have raised more money!

We’ve spent the past year developing new primitives for GPUs. We believe this enables new algorithms to change how machines learn, open new markets, and dramatically reshape AI’s cost structure. We have raised a $1 billion Series A, co-led by Index and Lux.

doubleAI avatar
⟳ Shift retweeted
doubleAI@doubleAI·Oct 19

What if code could think while it runs? Meet Maxwell: the world’s first Code Execution Model. A new species of AI that lives inside the machine. We believe it’s the key to unlocking 100% of compute. Requests for early access are open, link below.

1:41
Shift avatar
Shift@Shiftreduce·Oct 19

legends return

What if code could think while it runs? Introducing Maxwell: the world's first Code Execution Model. Maxwell lives inside programs as they run, and improves them! It's an entirely new class of model, sitting between code and hardware. No humans in the loop. IMO Maxwell is the most direct way to achieve superoptimisation at scale.

1:41
Off-By-One Conference avatar
⟳ Shift retweeted
Off-By-One Conference@offbyoneconf·Oct 12

📚 Slide decks from OB1 2025 & 2026 are live! Catch up on talks you missed, revisit top sessions, or dive into the technical breakdowns. Most slides are up on GitHub: 👉🏻 github.com/star-sg/OBO Huge thanks to all our speakers, sponsors, activity partners, and attendees!

Ronen Shustin avatar
⟳ Shift retweeted
Ronen Shustin@ronenshh·Oct 5

Haven't touched Android research in 5 years, but Android and I just became friends again 🤝. Reported a cool 3-vulnerability chain going from untrusted -> root. AI had a major impact on this chain, more info about that soon.

Post attachment