levonk

Father, technical leadership, AI, ML, automation, architecture, high scalability, fast data and big data. I keep busy!

[Mary Margaret Olohan @MaryMargOlohan]

NEW: Five FBI employees were fired today over the infamous Richmond Catholic memo on “radical traditionalist Catholics,” FBI source confirms to @realDailyWire.

[levonk @LevonK]

Library, video, medical, genetic, and financial history already have legal protections.

No AI history protections seems like a bad ruling:
https://t.co/msgM2CR7uj

[levonk @LevonK]

<3 @EFF's Right to Repair imitative. I'm wondering if as a stepping stone, an argument that mandated safety features that take a cheap sensor, add proprietary housing and software exploding it's cost is unfair. If the government mandates regulatory features the profit is a tax

[FBI Anchorage @FBIAnchorage]

🚨JUST IN🚨The Defense Criminal Investigative Service (DCIS), FBI Anchorage, and international partners disrupted four of the world’s largest Internet of Things (IoT) botnets that together were responsible for millions of infected devices and hundreds of thousands of DDoS attacks worldwide. 🔗https://t.co/SzcMSDAUdD
@USAO_AK | @DoD_IG

[levonk @LevonK]

@Steve_Yegge @voooooogel @Steve_Yegge sounds like we need a sovereign ?

[Adam Thierer @AdamThierer]

Another patchwork of bad AI laws coming. Politicians are about to get deeply involved in the news business in the name of saving journalism from AI. It's one thing to push transparency, it's another to micromanage all the varied uses of AI in the news business.

[levonk @LevonK]

Montana's newly passed "Right to Compute" law https://t.co/5m3MN9jwdb

@RightToCompute

What's everybody's take?

[levonk @LevonK]

Hollywood, @Disney, @warnerbros , @HBO , @paramountnet , etc... better have the visionaries in place before they become irrelevant. The trajectory the technology is on is exponentially improving. This will one day be something a teenager can do

[The Dor Brothers @thedorbrothers]

We just made a $200,000,000 AI movie in just one day.

Yes, this is 100% AI.

[levonk @LevonK]

@itamarsharify @windsurf Maybe this is only a Windsurf-Next thing?

[levonk @LevonK]

@windsurf Subagents & a TUI, pls?

[levonk @LevonK]

@WarrenInTheBuff The candidate uses a C bitwise optimization in TypeScript, likely handled by V8 already. In C, it shows deep understanding but risks premature optimization. In TypeScript, it's problematic.

[levonk @LevonK]

Some of you may have heard about the GLM 4.5, if you haven't it's worth taking a look based on the metrics, and my personal experience so far:

I thought I'd mention it, as they are giving away tokens on their discord by filling out a google form:
https://t.co/FgSU3IQGLL

[Elon Musk @elonmusk]

To figure out who will win any given technology race, just look at the rate of acceleration of innovation & growth

[levonk @LevonK]

One of the tools that has been really useful to me is @WisprFlow Dictating long prompts is so much faster than having to type it all out. If you want a month of Pro free (vs. the free basic account), you can use my referral code https://t.co/kIy9oBA6dV

[levonk @LevonK]

@itsandrewgao @FRobertsV hooks, Agents, sub-agents, https://t.co/95B2Ttpkm5

[levonk @LevonK]

@itsandrewgao - Agents + sub-agents
- Hooks
- The lack of global+project config, glob based application, and context size for those rules.
https://t.co/aEycBqaFnZ

[levonk @LevonK]

@dani_avila7 Haven't seen the behavior, so sorry if this is off track. I'm wondering if it's worthwhile to add an explanation of why you can't service them front-and-center, so they can direct the blame appropriately or give you feedback that there is enough mkt to justify dual deployments?

[Jackson Atkins @JacksonAtkinsX]

AI just unlocked 3x more power from GPUs.

A new AI framework called CUDA-L1 just taught itself to improve 250 different GPU tasks, delivering a 3.12x average speedup and a 120x peak gain.

Here's how it works:

The system's core is "Contrastive Reinforcement Learning (Contrastive-RL)," which is a leap over standard RL for code generation.

Standard RL is simple: the AI generates code, gets a performance score, and that score is used to update the model's weights. The AI never actually sees the score or reasons about it.

Contrastive-RL is different. The performance scores and previous code variants are fed back into the AI’s next prompt.

The model is forced to generate a "Performance Analysis" section by reasoning in natural language about why one version was faster. Then it creates an improved implementation.

The AI isn't just mindlessly iterating; it's performing a comparative analysis and building a mental model of what high-performance code looks like, allowing it to discover non-obvious strategies.

Why this matters:

Business Leaders: The 3.12x average speedup is a direct lever on your bottom line. This level of automation reduces GPU compute costs for both training and inference, freeing up capital and accelerating your product roadmap.

Practitioners: This isn't just a theoretical paper. The team open-sourced all 250 final, optimized CUDA kernels on GitHub. You can verify the performance gains on your own hardware (A100, H100, L40, etc.) today.

Researchers: This method provides a new blueprint for teaching LLMs to reason in specialized domains. The paper deep dives into "reward hacking" and how to prevent it.

AI is now building its own flywheel, learning how to maximize the resources we give it.

[levonk @LevonK]

@sdamico @ImpulseLabs_ Also, Any plans for a completely flat version so pans can slide around for cleaning, and one where a griddle can be evenly heated by close coils? Maybe make it so one of the induction coils moves (beneath the flat top) to allow griddle vs. multiple separate dutch ovens

[levonk @LevonK]

@sdamico @ImpulseLabs_ You and your team created something really impressive! A cool feature would be to use a powerful magnetic stirrer so things don’t burn. You’d have to make the stirrer spin and heat up simultaneously. Even have a reverse or sense it’s obstructed to maybe repel and retry

[levonk @LevonK]

@sdamico @mac01021 @ImpulseLabs_ 🤣