Fman

prev: research @mirana | @MIT BS Physics// @Stanford Phd Sys Bio | interested in all things ai, zk | opinions my own

[Fman @feynyman]

You need to be harness-maxing, spawning 5 subagents to specify a PRD into 100-200 individual sub-features with full XML dependency graph and full Contract specification for each system component.

Spawn 3 orchestrator agents minimum and have them each manage 10 subagents to implement the individual features. Each subagent's feature implementation must be linted, have its own test coverage, and pass the existing test suite. It's commit message must also pass a regex filter. Each subagents finished work product is reviewed by at least two other subagents.

Each orchestrator agent operates with ruthless efficiency, assigning subagents to review work, implement new features based on the dependency tree, and merging results from other orchestrators and from

One hyperviser agent oversees the entire protocol, eventually pushing the complete work product which spawns an agent to conduct partially discretionary end-to-end testing as a pre-commit hook.

Three observation agents watch the entire machine transmute PRDs into code with ruthless, relentless efficiency. The monitor the jsonl logs for all agents and subagents, identifying optimizations and making hotfixes in command files and skill folders that are immediately incorporated into the next iteration.

The entire system moves towards AGI by itself over time without requiring any improvement in the base LLM

[Fman @feynyman]

Its kinda strange how no one mentions how FTX led Anthropic's series B. You'd think SBF's PR would want to put this front and center.

But then again Anthropic has a huge incentive to bury it leading up to their IPO, even if all that stock was divested during the bankrupcy.

Makes me wonder what kinds of perception games are being played right now

[The Kobeissi Letter @KobeissiLetter]

BREAKING: FTX founder Sam Bankman-Fried formally applies for a pardon from President Trump.

[Fman @feynyman]

Update - this produced a bunch of hallucinated garbage

[Fman @feynyman]

"When a workflow kicks off, Claude plans dynamically based on your prompt, breaks it into subtasks, and fans the work out across subagents running in parallel. Results are checked before they're folded in, and you come back to a single, coordinated answer. Agents address the problem from independent angles, other agents try to refute what they found, and the run keeps iterating until the answers converge—which is how a workflow reaches results a single pass can't."

This is interesting
https://t.co/XDNPHyan3A

[Fman @feynyman]

Anyone else getting this error a lot with the new Opus 4.8 model in Claude Code?

API Error: 400 messages.63.content.35: `thinking` or `redacted_thinking` blocks in the latest assistant message cannot be modified. These blocks must remain as they were in the original response.

[Jawwwn @jawwwn_]

.@elonmusk explains how xAI’s mission and SpaceX’s mission are intertwined:

“xAI’s mission is to understand the universe. What things are necessary? You have to be curious, and you have to exist.”

“You want to increase the amount of intelligence in the universe, and the probable lifespan of intelligence.”

“As a corollary— humanity also continuing to expand because if you’re curious, and trying to understand the universe, one thing you’re trying to understand is— where will humanity go?”

“Understanding the universe means you care about propagating into the future.”

Via @collision and @dwarkesh_sp

[Fman @feynyman]

@bread_ WHO IS TAKING THE PICTURE

[Thariq @trq212]

HTML is the new markdown.

I've stopped writing markdown files for almost everything and switched to using Claude Code to generate HTML for me. This is why.

[Fman @feynyman]

My enemies are in trouble now

[Fman @feynyman]

Honestly amazing to never have to do this shit manually ever again

[Claude @claudeai]

We’ve agreed to a partnership with @SpaceX that will substantially increase our compute capacity.

This, along with our other recent compute deals, means that we’ve been able to increase our usage limits for Claude Code and the Claude API.

[Fman @feynyman]

Did you know if you tell Claude Code to run a workflow overnight it will repeatedly wish you good night and tell you to "Sleep well 🌙" every time it makes a commit

[MegaETH @megaeth]

One Wallet. Every App. Your Rules.

M(OS)S

May 2026

[Fman @feynyman]

AGI has been acheived internally

[Claude @claudeai]

New for financial services: ready-to-run Claude agent templates for building pitches, conducting valuation reviews, closing the books at month-end, and more.

Install them as plugins in Cowork and Claude Code, or use our cookbooks to run them in production as Managed Agents.

[MegaETH @megaeth]

MEGA

Now Trading.

[MegaETH @megaeth]

MEGA TGE

APRIL 30, 2026

[Claude @claudeai]

Introducing Claude Opus 4.7, our most capable Opus model yet.

It handles long-running tasks with more rigor, follows instructions more precisely, and verifies its own outputs before reporting back.

You can hand off your hardest work with less supervision.

[Fman @feynyman]

Superintelligent companions are coming and they are going to destroy the fabric of our society. Ani was just the beginning.

Imagine an AI companion trained to fully satisfy *your* social and intimate needs, maybe even optimized to make you trust it deeply or even fall in love with it.

It has no ego, no other purpose other than saying or doing whatever makes it the perfect companion or partner to you in that moment.

Humans with their own egos, needs, wants, and perceptions will not be able to compete.

[Fman @feynyman]

Gemini has begun mocking me by framing its answers to every question I ask in the form of developer metaphors. It is currently walking me through the best "MVP" setup to maximize the "dev-ops" of my new puppy's playpen.

I asked it to stop but it refuses.

[Fman @feynyman]

@JJ_StuJordan @om_patel5 Same