Ashish Verma

Say what you want about Rogan but you gotta admit it’s pretty impressive to be a spineless little bitch who stands for nothing

Never thought I'd see the day when I agree with Tucker Carlson, Megyn Kelly, Candace Owens, and Marjorie Taylor Greene... ... but they're ALL calling for the 25th Amendment to be invoked to remove trump from office. What a bizarre moment in time.

“HOW DARE YOU SPEAK LIKE THIS” -Tucker Carlson to Donald Trump One of the most prominent right-wing voices has broken away from Trump, directly criticising him.

@LeyvaCFM_ @grok is this video real or AI generated?

Oficialmente 🚨🚨 Elon Musk desafía a Trump y prohíbe oficialmente copiar enlaces X relacionados con la política Míralo tú mismo: Intenta copiar el enlace ahora

TWEEPS: We don’t do kings in America. 🇺🇸 We don’t do dictators. Tomorrow, March 28, show up at a No Kings march and remind Trump who this country belongs to: the people. I need 1,000 fast RTs and replies using #WeSayNoKings Please and thank you! 🙏💪

Pete Hegseth isn't just an unhinged and unfit drunk, it appears he is a raging racist asshole as well. As they prep to put 10,000 troops on the ground in Iran, he's blocking the promotions of highly regarded Black and woman officers. This is disgusting. 🤬

the @tplr_ai lore is epic: - Legendary Novelty Search when @const_reborn came in mid @MacrocosmosAI presentation pumped he got distributed training working: youtube.com/watch?v=UM5UhZ… (check around 32:00) - caused some fud as other open source research was used as inspiration... but that is literally the point of open source - @DistStateAndMe then anointed the leader of the project. realized immense work needed to be done. hired cracked team - epic research ensued: arxiv.org/pdf/2508.15706 - while other competitors had fundraising announcements, flashy UIs (with sometimes skeptical data being presented) and big PR campaigns, the team just kept building - Now by far the largest permissionless, distributed training run in history

𝗟𝗲𝗮𝗽 𝗳𝗿𝗼𝗺 𝗮 𝘀𝘂𝗰𝗰𝗲𝘀𝘀𝗳𝘂𝗹 𝗔𝗜 𝗽𝗶𝗹𝗼𝘁 𝘁𝗼 𝗮𝗻 𝗶𝗺𝗽𝗮𝗰𝘁𝗳𝘂𝗹 𝗲𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝘀𝘆𝘀𝘁𝗲𝗺 𝗿𝗲𝗾𝘂𝗶𝗿𝗲𝘀 𝗹𝗼𝗼𝗸𝗶𝗻𝗴 𝗯𝗲𝘆𝗼𝗻𝗱 𝗔𝗜 𝗮𝗽𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻 𝗹𝗮𝘆𝗲𝗿. Read more on post linked below

𝗬𝗼𝘂𝗿 𝗔𝗜 𝗮𝗴𝗲𝗻𝘁𝘀 𝗻𝗲𝗲𝗱 𝗯𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗰𝗼𝗻𝘁𝗲𝘅𝘁 𝘁𝗼 𝗯𝗲 𝗲𝗳𝗳𝗲𝗰𝘁𝗶𝘃𝗲. 🧠 Checkout details on my post linked below linkedin.com/posts/ashish-a… #MicrosoftFabric #AgenticAI #DataStrategy #DataEngineering #Ontology #AIStrategy #FabricIQ #AIEngineering

I’m losing respect for all the “experts“ that are surprised Gemini was trained on TPUs and not GPUs. Do you live in a cave?

@elonmusk Will robots be able to do ANYTHING a human can do? 🤔🤣

@_catwu Sandboxing will grow!!

We've shipped sandbox support in Claude Code CLI to make the CLI safer and faster. From internal usage, we've found that using the sandbox reduces permission prompts by 84% so that you can focus on the most important permissions

@_philschmid Hey i’m building a text-to-sandbox tool where you just tell it what you wanna try (poc, quick demo, learning something, whatever) and it sets up a clean sandbox for you in target cloud platform . No risk to you Looking for users to test it, pls dm if interested

Agents executing their own code is inevitable; it's too powerful not to happen. But I’m stuck on the architecture. Do we run this locally on the user's device, or safely in the cloud? Local sandboxes are the dream. Offline-first, zero latency, native access to your data. But can be easily done wrong. Allow Ingress/Egress? Access to environment variables? Permissions to filesystems, libraries? Remote feels more secure, easier to get started but much higher setup per individuall to access personal their data.

I personally would go with cloud based sandboxes with strong guardrails… with a local interface to it , something like vscode

💯 % agree…. Need isolated sandboxes that mirror prod envs, not just isolated runtimes, for reliable development and detailed logs for proper verifiability!

@karpathy we need decent sandboxes for verifying what agents do… most “sandboxes” today are just isolated runtimes. True verifiability needs the opposite: on-demand environments that feel exactly like production — networking, scale, latency, failures. That’s still the missing piece.

Sharing an interesting recent conversation on AI's impact on the economy. AI has been compared to various historical precedents: electricity, industrial revolution, etc., I think the strongest analogy is that of AI as a new computing paradigm (Software 2.0) because both are fundamentally about the automation of digital information processing. If you were to forecast the impact of computing on the job market in ~1980s, the most predictive feature of a task/job you'd look at is to what extent the algorithm of it is fixed, i.e. are you just mechanically transforming information according to rote, easy to specify rules (e.g. typing, bookkeeping, human calculators, etc.)? Back then, this was the class of programs that the computing capability of that era allowed us to write (by hand, manually). With AI now, we are able to write new programs that we could never hope to write by hand before. We do it by specifying objectives (e.g. classification accuracy, reward functions), and we search the program space via gradient descent to find neural networks that work well against that objective. This is my Software 2.0 blog post from a while ago. In this new programming paradigm then, the new most predictive feature to look at is verifiability. If a task/job is verifiable, then it is optimizable directly or via reinforcement learning, and a neural net can be trained to work extremely well. It's about to what extent an AI can "practice" something. The environment has to be resettable (you can start a new attempt), efficient (a lot attempts can be made), and rewardable (there is some automated process to reward any specific attempt that was made). The more a task/job is verifiable, the more amenable it is to automation in the new programming paradigm. If it is not verifiable, it has to fall out from neural net magic of generalization fingers crossed, or via weaker means like imitation. This is what's driving the "jagged" frontier of progress in LLMs. Tasks that are verifiable progress rapidly, including possibly beyond the ability of top experts (e.g. math, code, amount of time spent watching videos, anything that looks like puzzles with correct answers), while many others lag by comparison (creative, strategic, tasks that combine real-world knowledge, state, context and common sense). Software 1.0 easily automates what you can specify. Software 2.0 easily automates what you can verify.

🎯

Couldn’t agree more!

agent sandboxes give agents real workspaces: terminals, compilers, tests this allows them to code like engineers. it turns ai from code suggester into code executor. it’s also why ci/cd systems will soon treat agents as “non-human contributors”. the future of software development is human + agent commits, side by side.