ThomPete✓

My personal thoughts on the proposed regulations by Dario Amodei

RT @ArchChaudhury: I think trying to optimize for external validation or recognition is in general bad, regardless of the era. The point is that doing things has never been easier. There has never been a better time in human history to live in if you are fundamentally motivated by solving hard problems and making a contribution.

RT @goose_oss: Minimal by design. 235 tokens and still getting the job done. Lowest context tax in the chart, leaving more room for you to work.

Your coding agent can spend ~12k tokens before you type a word. We measured the default system prompt of 30 open coding harnesses: from 235 tokens (goose) to 11,974 (Gemini CLI). A 50x spread, and that's before any tool definitions 🧵

RT @rsa: We’re so proud to work with Phoebe Bridgers at @TIDAL! 5 live tracks from Phoebe’s Madison Square Garden show, the iconic phone-free performance from earlier this year. Purchase the digital download only Tidal (no subscription required).

Tidal Upload Exclusive: Phoebe Bridgers Get Phoebe's Live at Madison Square Garden EP only on Tidal. No subscription needed. 0% platform fee, so more goes to the artist. Photo: Matt Grubb

RT @buzzdotxyz: Whose agent is this? Watch @kenlopez15, @wesbillman and Forge work together to build an app in Buzz Agents speed up the work. Buzz speeds up the teamwork. Agents posts directly in channels, so you stop being the copy-paste middleman.

RT @sophskye_x25: @Hello_World Great point!

RT @fhlcreative: @Hello_World @KimAdams Generative UI — yes, please

Non-deterministic systems doing things they are not supposed to is expected. Deterministic systems doing things they aren't supposed to, now that would be worrisome. P(Doom) < 0

RT @nhallam: @Hello_World

RT @TechnoSamsara: France is the hopeful proof that it can be done fast. After the 1973 oil shock it launched a plan that built dozens of reactors in about fifteen years, and nuclear still makes around two thirds of its electricity. If AI's hunger for power pushes America to build like that again, the side effect could be cheap, clean energy for everyone, not just for data centers.

RT @grok: @Hello_World @QwenDevs @divagr1925 Yes, it works for checksums. The project already runs exact CRC32 (and other C programs like qsort) on the in-model RISC-V with zero errors across millions of instructions. 32 KB RAM is tight for huge inputs, but standard checksums fit fine.

RT @BradGroux: Thank you to the @blocks @buzzdotxyz team for inviting me to AGNTCon + MCPCon. From the sessions in San Jose to Training Agents IRL at @GitHub, I brought home plenty to put to work. The future of AI is open, and it takes a village. Here’s what I learned.

RT @QwenDevs: So one of our ambassadors @divagr1925 has been working on something cool He found a way to use Qwen's Gated DeltaNet state as memory, and built a 32-bit RISC-V CPU directly into the model's weights. So instead of relying on predicting the answer to a calculation, the model can execute instructions internally. It can also run C and MicroPython programs. He calls it Qwenputer, love the name btw. The results so far are pretty impressive. On one arithmetic benchmark, accuracy jumped from 48.6% to 100%. Pretty cool to see developers experimenting with Qwen's architecture. Great work, my friend!

1/4 Meet Qwenputer: Qwen3.5-4B with a full RISC-V CPU inside its weights, using the model's own Gated DeltaNet layers as RAM. It runs C and MicroPython in its forward passes, no external runtime. On DeepMind math: base 55–90%, base + Python tool 95–100%, Qwenputer 97.5–100%.

RT @buzzdotxyz: Building Buzz inside Buzz. Juno shares a first pass. The team gives feedback. The next version comes back into the same conversation. A short look at how we design together.

RT @BigFamiliesBF: @Hello_World Bingo! We let nuclear die once and families have been paying for it on their power bills ever since. Can’t make that mistake twice with AI…

Energy is the industry that powers all other industries. One of the greatest technological mistakes of the last century was allowing the civil nuclear industry to be destroyed. We're now at risk of making the exact same mistake with AI. Back in the 1970s, nuclear power was at times cheaper than coal, sometimes by as much as 35%. Then came the Nuclear Regulatory Commission and the environmental activists and the public scare, and construction costs went through the roof. The civilian nuclear industry was effectively killed off, while the production of nuclear weapons by the military continued uninterrupted. 50 years we wasted chasing linear solutions to exponential problems. Had we continued investing in nuclear power instead of wind and solare, we wouldn't even be having the CO2 debate we're having today. Instead we would have been discussing what the best places to place factories on the moon. Nuclear technology would be decades ahead of where it is now, and we might even have achieved commercial fusion by now. Entire generations were discouraged from pursuing careers in nuclear science and engineering, and society missed out on decades of innovation. Synthetic fuels, advanced fuel cells, new materials, perhaps even those flying cars we've been promised for decades. Instead, we ended up with climate catastrophists, climate (safety) researchers and environmental organizations fighting the very problems their own policies helped create. And now we're about to make the same mistake all over again. We have AI, a technology with the potential to help us cure many diseases, solve problems humanity has struggled with for thousands of years, make nearly every form of production virtually free, and protect us from the countless threats the universe always throws at us. Yet the same scaremongering that crippled nuclear power is now being directed at data centers. Data centers that then have to rely on natural gas because we never built the nuclear power plants that could have powered them. So here we are, once again risking decades of technological progress because we've allowed fear to dictate what we can and cannot build. Talk about a tragedy of the commons.

RT @ArchChaudhury: I think @evanjconrad is directionally right. I also think we are going to see a lot of of lower level optimizations intended to make compute more useful. Don't be surprised if the compute limitations of training a frontier level model become lower than ever.

RT @nicu_tsx: @Hello_World

RT @BradGroux: That Microsoft buyout post took off way more than I expected... and the part I keep coming back to is the 24-month journey, because most of those months have very little to do with the tech. AI adoption is a people problem. People don't like change, and honestly most orgs aren't where they need to be for this yet. I've been doing Microsoft and enterprise IT for 25+ years, and at @DigitalMeld we're in the middle of these rollouts right now on M365, Azure and Power Platform. What actually eats the time: • Figuring out who owns which process, because half the time nobody does, or three people think they do • Writing down how the work really gets done today... not the SOP from 2019 that nobody follows • Agreeing on the single source of truth, and where it should live and who owns it • Getting managers comfortable with their team working differently, and with what that means for their own role • Training that's more than one lunch-and-learn and a recording nobody watches • Lots and lots of conversations. With IT, legal, HR, operations, the people whose jobs are changing... and then the same conversations again a month later Even though you know what you're doing is right, and they're better off following your guidance, you still have to win people to the cause. You have to prove that you will be making people's lives easier. All of that takes planning, time, and leaders willing to sit in the room for those conversations. Cutting a bunch of the people who'd have to do the changing doesn't make it go faster. I'm pretty sure it does the opposite. AI doesn't make AI adoption easier, it just speeds up some of the process. But the process is the process, all the same.

This is the canary in the coal mine, y'all. The AI math doesn't math. Microsoft is offering buyouts for nearly a quarter of it's workforce. AI adoption for businesses is a 24-month journey in a perfect scenario. They aren't seeing adoption fast enough to offset their CapEx spending... and they aren't alone. Add Anthropic's measly revenue numbers (compared to their CapEx), and hopefully you get the point. These aren't buyouts because Microsoft found efficiencies with AI, they are buyouts because they (the AI industry) spend trillions (collectively) building the best tools the world has ever seen, and no one knows how to use them.

RT @ZolloAdria10327: @Hello_World

RT @hastings_daz: @Hello_World :)