Zengineering

My curated list of good follows for GenAI / ML twitter.com/i/lists/165767…

@policytensor China can’t alienate other suppliers in the GCC. They won’t step in that way

@Samhanknr The quickest way is to have China sign on to a unbounded weapons resupply agreement on credit/paid by the aggressors. But there may be other ways. They are just much more difficult.

What are the secret clauses? None of this offers any insurance that Iran is not be attacked again by the aggressors. The whole proposal seems premised on the idea that Iran wants economic relief more than it wants fundamental security. It’s a nonstarter.

Most people are sleeping on the fact that Copilot CLI is just better than Claude Code or Codex subscription. 1. You switch between codex and claude in same subscription 2. $39 sub is great value (3-4x usage than $20 plans) 3. the CLI is so good! no flicker and /slash commands work even when model is working. Unlike CC or Codex.

advice for the post AI world

Vishwaguru

@yoavgo Just use Claude code to make web pages

the things that keeps me fascinated about AI is how odd some of the tradeoffs are. I mean who would have thought that creating presentations by generating images and editing the content with prompts that transform images would be easier for google to develop than something that actually creates and edits structured content.

@GeorgeMayer ChatGPT it

@Samhanknr You want me to read this whole thing lol?

One interesting combination of writing software that uses LLMs with LLMs is the resulting pastiche of ideas written, tested, succeeded and failed and the permutations therein. A mix of tools, prompts, context management, infra. Unlike previous eras where with enough planning you could know whether something is feasible, building around LLMs must be tested/benchmarked. You kind of need to prototype. Your intuition builds over time, but you still don’t know if something will work and improve your system. This creates a mess even if managed well and agents struggle to reason over it and erode their natural disposition is to add not remove. This type of software needs clear designations around what is canonical and what is experimental, maybe this is just the role for flags and docs, but feels like something is missing.

@GeorgeMayer This is essentially the state of most machine learning code

@GeorgeMayer You can look at repos like Nano GPT speed run as they run into very similar problems

@badlogicgames @abhimanyulad This is less common for devs from India to have such a diversity of careers.

@Samhanknr @abhimanyulad many things, mostly during summers. - industrial welding - landscaping and lots of tractoring - construction sites - production line q&a - waiter and a bunch of other shit i forgot.

i love this comment. it illustrates how we software people permantently get real-life wrong, then build systems that are supposedly helpful, when they are not. bonus points if you claim the system replaces humans. the world is more complex than it seems. domain expertise is still everything.

There will be more of this. And as much as we're joking about it, we're seeing a massive degradation of code quality right now and we're increasingly only catching it way too late.

@badlogicgames @abhimanyulad What did you do before SWE

@abhimanyulad it does make me wonder, how many SWEs have had normal jobs before they became SWEs.

@championswimmer How do you know it’s actually Boris and not a skill.md 😛

Agentic this, agentic that. And yet..... Claude Code customer support is tagging Boris and Thariq on Twitter. What is Anthropic (and their customer support partners Intercom) doing about this?

Deal of the Day March 23: New MEAP! HALF OFF my book Evaluation and Alignment, The Seminal Papers, and other selected titles @ManningBooks hubs.la/Q03-d27Y0

Introducing Hyperagents: an AI system that not only improves at solving tasks, but also improves how it improves itself. The Darwin Gödel Machine (DGM) demonstrated that open-ended self-improvement is possible by iteratively generating and evaluating improved agents, yet it relies on a key assumption: that improvements in task performance (e.g., coding ability) translate into improvements in the self-improvement process itself. This alignment holds in coding, where both evaluation and modification are expressed in the same domain, but breaks down more generally. As a result, prior systems remain constrained by fixed, handcrafted meta-level procedures that do not themselves evolve. We introduce Hyperagents – self-referential agents that can modify both their task-solving behavior and the process that generates future improvements. This enables what we call metacognitive self-modification: learning not just to perform better, but to improve at improving. We instantiate this framework as DGM-Hyperagents (DGM-H), an extension of the DGM in which both task-solving behavior and the self-improvement procedure are editable and subject to evolution. Across diverse domains (coding, paper review, robotics reward design, and Olympiad-level math solution grading), hyperagents enable continuous performance improvements over time and outperform baselines without self-improvement or open-ended exploration, as well as prior self-improving systems (including DGM). DGM-H also improves the process by which new agents are generated (e.g. persistent memory, performance tracking), and these meta-level improvements transfer across domains and accumulate across runs. This work was done during my internship at Meta (@AIatMeta), in collaboration with Bingchen Zhao (@BingchenZhao), Wannan Yang (@winnieyangwn), Jakob Foerster (@j_foerst), Jeff Clune (@jeffclune), Minqi Jiang (@MinqiJiang), Sam Devlin (@smdvln), and Tatiana Shavrina (@rybolos).

@championswimmer Software engineering - Coming soon in H2 2026

Truth is, there is no way to make the existing OpenClaw beast secure If you want a secure agent you have to build it differently from ground up. Stop taking shortcuts. Pete didn't make this with security as his goal. He made it for tinkering.

We need a name for performative agent parallelization I propose: “slop theater”

@championswimmer If you want a high standard of living it’s competitive everywhere

@championswimmer There’s no place on earth where moving there will solve all your problems 😆

Shobhit moderating two people both saying grass is greener on the other side. Lol. Person who stayed in India asking people to live. Person who went abroad saying there are no opportunities. Absolute cinema.