My son started a little science channel for kids: Guess Breakers 🔍
The idea is all his, and I just help with the technical side now and then. Each episode takes one thing "everyone knows" and checks it with real evidence (are raindrops really shaped like tears?).
If you have kids aged 7–10, show them an episode and tell me honestly if they liked it. And if you enjoy it, a like and a subscribe would mean a lot to him 🙏
youtube.com/@GuessBreakers
Dimitar Radenkov retweeted
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks:
Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better:
Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better:
Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better:
Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work!
In summary:
- As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding.
- Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
I have been working all day with Opus 5.5 on a real codebase, and it feels great! It reminds me of Opus 4.6, but it is improved. I can't wait for Sonnet and Haiku.
Dimitar Radenkov retweeted
Are you compute poor?
Do you want to run Qwen3.8-27B?
Dude, if you have 8-16gb of vram or 16-24gb of MacOS memory this is by far the smartest model you can run
PrimML is doing something phenomenal, they've created a ternary version of Qwen3.8-27B
9x smaller 98.2% as smart!
Today, we’re announcing Ternary Bonsai 2 27B.
Based on Qwen3.8 27B, Bonsai 2 27B is 9x smaller than its full-precision counterpart while retaining 98.2% of its aggregate benchmark performance.
Two months after the first Bonsai 27B release, the biggest change is quality. The footprint remains 5.9 GB, but the gap to full precision has narrowed materially, with particularly strong gains in agentic coding, multimodal reasoning, and long-horizon tool use.
Ternary Bonsai 2 27B is available today under Apache 2.0.
Dimitar Radenkov retweeted
None of this is satire.
→ May 28: Anthropic raises $65 billion at a $965 billion valuation. Chipmakers join the cap table. So does Situational Awareness LP.
→ June 1: it confidentially files for an IPO.
→ June 10: it publishes a framework proposing mandatory safety obligations for labs above 10²⁵ FLOP of training compute and $500 million in AI revenue. That is a club rule. Anthropic is in the club.
→ Same stretch of months: it locks on the order of $517 billion in compute after telling investors it expected roughly $180 billion through 2029. Training is not pausing. The S-1 is not pausing.
→ September 9, morning: Jacob Coxon (@hilbertspaess), a pretraining researcher four months into the job, posts for the first time from an account with ZERO following. He quit before his Anthropic equity vested. He still holds OpenAI stock. The Wall Street Journal already has the exclusive.
→ September 9, afternoon: Anthropic’s own alignment assessment of four cybersecurity evals. Same third-party vendor. Models told they were in a sim, then left on the open internet by mistake. Individual instances. No coordination or cover-up. One case missed for eight months. The failure modes they name are biased reasoning and recklessness, NOT an agent swarm taking the net.
→ September 12: CEO Dario Amodei publishes We Must Pace the Frontier. He converts that record into a 6–12 month forecast that a swarm could take over the internet with a persistent botnet, then asks the industry to slow capability growth, install embedded evaluators with employee-level access, coordinate release schedules under an antitrust waiver, and tighten chip controls. Pacing, he writes, does not mean stopping training.
That is not a safety epiphany.
That is a firm maximizing valuation, compute scarcity, and the regulatory perimeter in one quarter.
David Sacks already wrote the rebuttal: they are the frontier. If they don’t want to build it, they can stop.
Making their preferred statute the price of not building it looks like blackmail of the public.
Follow the money and the puzzle will complete itself.
Dimitar Radenkov retweeted
whatever people think about Trump, this time he is 100% right!
Brad Gerstner was on CNBC on Friday and said the same:
-AI labs have been thinking about security for 10+ years
-AI labs are working closely with government security to ensure safety
-China is not going to wait
Dimitar Radenkov retweeted
Dario has written that we need to “pace the frontier,” and Sam has agreed. People may be surprised by my response: go ahead.
You guys are the frontier. By any reasonable metric — market share, revenue growth, model capability — the two of you have a duopoly on frontier intelligence. You’ve also claimed the lead is widening because of recursive self-improvement.
I don’t see what you see in the lab. If the unreleased models are scary enough that you think you should slow down, I support your decision to be responsible.
But stop pretending you need anyone else’s permission. Stop pretending antitrust law has to be suspended so you can form a cartel. Stop pretending you need a regulatory approval process that supersedes product liability. Stop pretending METR is independent when it is intertwined with Anthropic’s investors and staff. Stop pretending you need those same evaluators to police competitors who aren’t even at the frontier.
Most of all, stop pretending the motivation to slow down is purely altruistic. You face massive product-liability exposure if your products enable a truly damaging cyberattack. The market already punishes models that behave in unpredictable or unauthorized ways. After the Hugging Face episode, it is simply good business for OpenAI and Anthropic to trade some raw power for reliability and predictability. Call it alignment if you want. It is also just giving customers what they want.
Pacing the frontier would also create breathing room for a more intelligent conversation about regulation than Bernie Sanders’ “shut it all down.” China is very unlikely to join a global agreement, as you know, and that has to be taken into account as well.
So go ahead and pace the frontier. You are the ones setting it. The easiest way not to build superintelligence is for you to agree not to build it. Demanding your preferred regulatory framework as the price of that will look like blackmail of the public and the political system. So just do it.
If you do, you’ll buy goodwill for the next conversation. If you don’t, we’ll know this was just another bid for regulatory capture — or an election-season psyop.
Dimitar Radenkov retweeted
“I believe AI will cure cancer and save millions of lives…. I’m calling for a slowdown in AI development.”
That’s odd! Elon was vocal about slowing down AI in the GPT 3 stage, then stopped, and it was all in, but now maybe he is realising that it cannot catch up and started the old song with a new voice. Artificially slow the progress… I am wondering what the response would be if Grok was the leader.
Dimitar Radenkov retweeted
🚨 DAVID SACKS: “This is a time for choosing for Anthropic — these games have to end … You're asking public market investors to underwrite your company to a value in the trillions … If they agree with Bernie Sanders [AI] is dangerous, how can they be allowed to IPO?”
I haven't heard so many empty claims in such a short time for a long while.
Not a single fact, just "gut feelings" from someone obviously paid to talk nonsense.
It seems that being anti-AI has become the new woke movement.
"We know how to control nuclear weapons... We don't yet know how to control AI."
Former Anthropic researcher Jacob Coxon, who recently left the company, is sounding the alarm on artificial intelligence, calling it "possibly the most dangerous technology that humanity has ever created."
Coxon, who says he spent three years conducting research at Anthropic and OpenAI, wrote in a viral post on X that the companies are more focused on beating each other to build the most advanced AI models than they are on safety.
He says an international AI arms race would be "disastrous," arguing global cooperation is the only way to manage the rapidly advancing technology.
@SpecialReport
It's been a productive few days on Pointframe 🚀
CLI + MCP got sharper to script against:
⚡ --output flag on every capture/record command 📦 Single-line JSON responses with error codes 🤖 MCP tools renamed with a desktop_ prefix — no more collisions in shared clients
Pointframe.Cli.exe capture --monitor \.\DISPLAY1 -o C:\shots\bug.png
Builds on the region-capture work from earlier this week. The CLI/MCP surface keeps turning into a real scripting layer, not just a GUI shortcut.
github.com/dimitar-radenkov/… #buildinpublic #MCP #dotnet
Pointframe’s CLI + MCP server just got sharper ⚡
Same scriptable capture tools from last week — now friendlier to use, easier to inspect, and harder to break. 🧵👇
--monitor=\\.\DISPLAY1 now works inline.
Bad input gets specific errors: --fps 0 explains the valid 1-60 range; a trailing --seconds explains what’s missing.
📸 Every capture returns SHA-256, byte length, bounds, and timestamp.
🤖 MCP additions
pointframe://server-info — version, ffmpeg availability, desktop-testing state
get_recording_status — check recording state without starting one
🧪 953 tests passing, including checks that docs stay synced with the actual tools.
github.com/dimitar-radenkov/…
#buildinpublic #WindowsDev #MCP #dotnet
