Co-Operability @Cooperability

Gradient descent for SKILL.md files sounds interesting, maybe a bit complex but it's becoming a real part of agent harness. SkillOpt is one of the first papers to treat markdown skill files as trainable parameters and provides a proper optimization framework for them. A few things I learned that you should consider too. 1. The validation gate is the only thing that matters in a self-editing loop. Held-out set, strict improvement, ties rejected. End-to-end, their best skills land with 1 to 4 accepted edits total. If your "self-improving agent" is accepting most of what it proposes, you're shipping slop. 2. Bounded edits are better than full rewrites. 4 to 8 edits per step is the sweet spot. Remove the budget and performance collapses. This is the textual analog of learning rate, and it transfers to any LLM-as-author loop. If you're using an agent to refactor your docs, your prompts, or your skills, cap the diff size. 3. Compactness wins. Median final skill: ~920 tokens. Skills do not need to be long. They need to be high-signal. Most skill files I see are bloated because length feels like effort. It isn't. 4. The harness is becoming less important; the skill is becoming more important. A Codex-trained skill ported into Claude Code hit +59.7 points on SpreadsheetBench. Procedural knowledge is more general than the runtime that produced it. 5. Frozen model + trained context is the practical adaptation. GPT-5.4-nano with a SkillOpt'd skill ≈ frontier behavior on procedural benchmarks. Cheaper, portable, inspectable, zero inference-time cost. This is the answer to "how do we adapt a frontier model for our domain" for almost everyone who isn't training their own models. 6. Verification is the bottleneck. Every gate in this paper depends on an auto-grader. That works for benchmarks. It fails for writing, design, and strategy, exactly the open-ended work we want to automate. Whoever builds the verifier for open-ended tasks owns the next stage. There are also two leassons I learned while shipping v2.3.0 of my Context Engineering Agent Skills repo, measured across composer-2, claude-opus-4-7, gpt-5.5, and gemini-3.1-pro via the @cursor_ai SDK: - Description and body are two different surfaces. The router only sees the description. The agent sees the body once activated. They can quietly disagree, and only end-to-end task tests catch it. - Aggregate accuracy is the wrong unit. When I rewrote three descriptions, the corpus average moved ~1pp. Individual skills moved 23–25pp. Per-skill effect size is where the action is. Also, in Feb 2026 I shared a piece called Personal Brain OS arguing that the markdown file is a first-class substrate for agent state. SkillOpt is the optimizer-shaped version of that same argument: not "store memory in files" but "treat files as trainable parameters with proper optimization machinery around them." That's the move from static to measured. The fast/slow split they describe already lives implicitly in the digital-brain-skill repo: - voice-guide and tone-of-voice.md are slow-state (rarely touched) - posts.jsonl and bookmarks.jsonl are fast-state What SkillOpt adds that I didn't have is a protected section invariant, a structural guarantee that fast edits cannot overwrite slow lessons. Removing that mechanism cost them 22 points on SpreadsheetBench. Worth borrowing. If you're building agents, SkillOpt: Executive Strategy for Self-Evolving Agent Skills is a good paper to read: arxiv.org/pdf/2605.23904

Healthcare jobs accounted for 77% of all the new jobs created over the last two years

You can't automate what you don't understand.

+1 for "context engineering" over "prompt engineering". People associate prompts with short task descriptions you'd give an LLM in your day-to-day use. When in every industrial-strength LLM app, context engineering is the delicate art and science of filling the context window with just the right information for the next step. Science because doing this right involves task descriptions and explanations, few shot examples, RAG, related (possibly multimodal) data, tools, state and history, compacting... Too little or of the wrong form and the LLM doesn't have the right context for optimal performance. Too much or too irrelevant and the LLM costs might go up and performance might come down. Doing this well is highly non-trivial. And art because of the guiding intuition around LLM psychology of people spirits. On top of context engineering itself, an LLM app has to: - break up problems just right into control flows - pack the context windows just right - dispatch calls to LLMs of the right kind and capability - handle generation-verification UIUX flows - a lot more - guardrails, security, evals, parallelism, prefetching, ... So context engineering is just one small piece of an emerging thick layer of non-trivial software that coordinates individual LLM calls (and a lot more) into full LLM apps. The term "ChatGPT wrapper" is tired and really, really wrong.

tobi lutke @tobi

a year ago

I really like the term “context engineering” over prompt engineering. It describes the core skill better: the art of providing all the context for the task to be plausibly solvable by the LLM.

346 879 9K 2.1M 2K

In which the head of a 20,000-person internal security agency, whose members are often masked, heavily armed, and operate out of unmarked vehicles, ‘demands’ an end to political speech critical of their operations. There is a word for this.

U.S. Immigration and Customs Enforcement @ICEgov

a year ago

📺ICE acting director responds to political rhetoric causing anti-ICE sentiment and threats to federal officers and their families: “I’m not asking them to stop — I’m demanding that they stop.”

5K 3K 12K 4.3M 789

We tested every major AI reasoning system. There is no clear winner. Accuracy goes up as you stack modern CoT techniques, but efficiency goes way down. This gives rise to a Pareto frontier on accuracy vs. cost using ARC-AGI as a consistent measuring stick.

When Claude instances talk to each other, in ~90% of open-ended interactions they spiral into discussions of consciousness, then profuse gratitude, then abstract spiritual/poetic expressions with Sanskrit and emojis.

"Leadership skill with AI agents predicts leadership skill with human groups." Very proud to have contributed software to the underlying stack that helped conduct these experiments. AI/Human leadership performance was strongly correlated (ρ=0.81)! Paper: nber.org/papers/w33662

It doesn't matter that OpenAI has 400 million "weekly users" - they spent $9bn in 2024 to *lose $5bn* (set to double in 2025), their products are either commoditized or do not work, and they'd die quickly without VC welfare. They're not a real company. wheresyoured.at/wheres-the-mon…

keyboard_arrow_left Previous keyboard_arrow_right Next

Three years ago today, the military forces of Russia launched an attack across the borders of #Ukraine on the ground and from the air. As Russian troops drove down roads in the hope of a short war and a victory parade in Kyiv, they received a mighty surprise. 1/16 🧵🇺🇦

It is grimly hilarious to see this guy ripping up decades of American soft power projects while going “Wow!” and “Gawrsh!” at stuff he doesn’t understand. God only knows what the Russians or Iranians are making of the CIA getting absolutely hosed in public by an imbecile.

The anger over Deepseek training on ChatGPT's outputs, in violation of TOS, is justified. But it might perhaps also make the AI companies think about the training they did on copyrighted content without consent. techcrunch.com/2024/12/27/why…

Mint it, Print it, Sell it Fuck it. That is the mantra for the world you are suggesting. This is the biggest bunch of self serving Bullshit I have ever heard. In your world there is no ownership. Just speculation. Hello every scam targeted at everyone and anyone who has no clue about crypto Good bye whatever hope the crypto industry had of legitimizing itself @garygensler has got to be laughing his ass off. And the new head of the SEC has got to be pissed. If he believes this nonsense he has no ability to enforce any SEC law. Stock, token. Bond. Nothing Rather than modifying the IPO process to accept crypto and give investors some understanding of what they are buying . Any registration requirements just flew out the window. I’ll say it again. Mint it, print it, sell it Fuck it. That is the mantra for the world you are suggesting.

Jeff Dorman @jdorman81

2 years ago

The market is losing its mind over the $TRUMP coin, and completely missing the plot. Here’s why this is going to be incredibly long-term bullish for the industry (and it has nothing to do with TRUMP coin itself). Thread 👇

571 1K 8K 6.4M 3K

“The $TRUMP memecoin — a financial asset that didn't exist on Friday afternoon — now accounts for about 89% of Donald Trump's net worth.” Trump just became one of the top 25 richest people on earth. Through this. Insane, open & now normalized corruption. axios.com/2025/01/19/don…

Pretty obvious Trump is talking to nobody about this / some scammer is running the show I don’t think coinbase + binance sign off on spot listings if they knew another coin was coming within 24 hours

@darkpools @MELANIATRUMP single holder owns 89% that's all you need to know for melania

I've been addicted to cultural commentary about America because it's so current, urgent and fascinating. This sprawling coverage of American solitude delves into partisanship, economics, rural/urban history, gender roles, and religion. Worth your time. theatlantic.com/magazine/archi…

Newsletter: OpenAI is a bad, unprofitable business, with most of its revenue coming from premium subscriptions, and only 27% ($1bn) coming from people licensing its technology, suggesting that the generative AI industry is much smaller than we thought. wheresyoured.at/oai-business/

It is impossible to overstate how despicable and dangerous Trump's demagoguery is, even if he loses again. It’s blame America first. He surpasses even the far left in calling America a bad actor, irreparably broken, a hellhole that only he can save you from. He’s the anti-Reagan.

Ron Filipkowski @RonFilipkowski

2 years ago

Trump: America is at fault for Russia’s invasion of Ukraine, not Putin.

2K 2K 7K 3.6M 732

Elon: "I've kept lying to you again and again about Tesla's 'Full Self Driving' capabilities for the last 8 years, but you should believe everything I say about politics and everything else." x.com/RealDanODowd/s…

Dan O'Dowd @RealDanODowd

2 years ago

x.com/i/article/1834…

136 254 2K 718K 200