HNHacker News
TopNewBestAskShowJobs

lubujackson

5,379 karma · joined October 1, 2010

https://yagmin.com/ https://github.com/yagmin/ https://linkedin.com/in/jyagmin/
submissionscomments
lubujackson··on Use Task Runners for Common Coding Tasks
Oo, I like that header font: https://fonts.google.com/specimen/Germania+One?preview.scrip...
lubujackson··on The Greenhouse and the Lens: Two Modes of Agentic AI Work
Let's assume advances in harnesses, models and efficiency continue to rapidly improve. Let's say we only run on solar power.

The concept of "waste" completely shifts. Like in 1990, I had to decide if dow loading an image was worth it, because it would take 5 minutes or more. Now, it isn't even a remote concern, even though images are clearly "wasting" bandwidth sending a 12mb image to my tiny phone screen.

It is much more interesting to think about the cultural impact of that shift. You can be concerned about the impact of AI on various fronts, but worrying about "waste" is tilting at windmills.

lubujackson··on Amazon spent $1.8M using Claude for menial coding task, went 860% over budget
Some schadenfraude because of AWS's lack of a hard spend cap, which has screwed countless little fish and individuals over the years.
lubujackson··on Google cancels AI Studio app after 800k preorders
Giogle sunsetting has reached its final form...
lubujackson··on The Silicon Valley Founder Meat Grinder
I assume it is like the 70s wood-paneling - signifying some vague commercial sense of ruggedness and nature while completely missing the mark.
lubujackson··on I flagged two research papers for fake authors and both were accepted as orals
The entire concept of publishing as a validation step for actual science has been declining for decades, fully co-opted by pay to play and virtue signaling for academic hiring.

AI is simply going to bury it as a useful system. What will grow from the ashes will be something much more dynamic, where verifiable data is the gold and the conclusions and associated details will persist only as the human-level translation.

lubujackson··on The End of an Era
It is just like coding.

If you are on top of the AI, it is a powerful tool for exploration of ideas, organization of thoughts, bridging connections, outlining, filling in gaps.

But if you vibe write, it will look good on the surface but be full of illogical, contradictory statements, disconnected characterizations, etc.

lubujackson··on The End of an Era
This is the black and white perspective that literally every other human-creative industry is wringing its hands over. But I think it misses a key difference that is already painfully obvious, which is creating new and engaging things will never be as simple as pressing a button. There will always be a human at the other end, pressing the button (or a human who deploys the agent, etc.)

The best comparison is photography. Instantly, a whole career path of portraiture was tossed out the window. Go to a museum and you see portrait after portrait - the best ones. There were millions of shitty ones made at all levels of society, and that job was replaced by cameras. But painting didn't go away, it got weird, it became anti-camera and more human, in a way.

This splitting of purpose and amplification of humanness is going to ramp up. Expect everyone to grow a taste for weirder, unprocessed music, experimental novels, wild and unexpected designs. Art is human communication and AI brings that into sharp relief.

Novels seem the most protected from AI - pulp/romance/beach novels will be crushed by volume and the sad remains of the publishing industry along with it, but interesting human novels will be just fine. And, much like with photography, I think AI will eventually normalize to its own artistic medium with writing, where careful usage will allow more ambitious work to appear that wasn't possible before, like a novel written using only sestinas, for example.

lubujackson··on After the AI Crash
Be careful if you expect a dot com fallout - that was retail investor driven and took a long time to unwind as people sat through painful drops. Ai is much more of a private investment bubble. AI remains useful. What is likely to go away (and all at once) is investment and free rides/discounts.

So I expect more of a sobering process for AI companies rather than a blowup, simply because they all still will have cash in the bank and some have actual products with clients that make use of them.

lubujackson··on Paged Out #9 [pdf]
I've never seen this before, it's like a modern 2600 - deeply technical, scattered topics and depth, all very hacker-curious - and beautifully designed to boot.
lubujackson··on Cursor Bridge – Run Unlimited Claude Code on Your Cursor Subscription
Yes. Cursor is significantly better at understanding context without needing a 5 page spec or burning a truckload of tokens. And it's debug mode is top tier if you haven't tried it - people sleep on Cursor.
lubujackson··on The AI jobs apocalypse probably isn't coming anytime soon
Depend what your company actual does. More relevant tasks I have seen involve 2D arrays or transforming data from a Python dict, or parsing JSON. The idea is to test the ability to understand and manipulate complex objects, which is a good generic bar.
lubujackson··on The AI Productivity Illusion
Insightful point, but we should also consider we are in a very transitory period where were are adjusting every level of these tools (how verbose? How sycophantic? What should it ignore, which details matter, etc.) And more importantly, we as users are becoming attuned to reading and gaining understanding with AI.

When everyone first uses AI they are swayed by its intelligent sounding sentence construction. It took time to be able to look past that and understand basic assuptions still needed to be questioned.

lubujackson··on The Dark Night of Mathematics
I think the hand-wringing is a bit premature.

Some mathematicians (like SWEs) view AI as a boost to their ability to explore and understand possibilities - it is a great tool. Others see it as taking the soul out of what they do (loss of flow or craft). Most worry about losing their jobs.

I think for both mathematicians and SWEs, there is real loss for how we worked in the past, but I don't think job loss will be quite as severe as feared (and may actually increase). Because we still need, more than ever, people who understand the output and, more importantly, can ask the right questions.

lubujackson··on The new rules of context engineering for Claude 5 generation models
Recalling "They're Made of Weights"
lubujackson··on Ask HN: How do I find startups that are looking for a dev? How to approach them?
Wellfound.com, workatastartup.com
lubujackson··on Long presumed dead, a thriving coral reef is discovered in West Africa
There are 6 floristic kingdoms in the world, and Fynbos can only be found in a tiny section of Africa.

When I was there it was wild to see a completely different array of plants and flowers. There is one small white flower that had petals that felt like some sort of sythentic material. There is a whole world that very few people experience or even know exists.

lubujackson··on "Drawing" the Mona Lisa with GPT-5.6, Claude, Gemini, and Grok
I do think they are going to stretch their lead in value if Anthropic doesn't wake up and stop YOLOing tokens. Kimi is an amazing achievement, but it has the same (or worse) kitchen sink approach as Fable.

At work, even if Fable is technically better I much prefer Sol because it is so much faster and concise.

lubujackson··on A digestion of the Jacobian conjecture counterexample
I am sure he did the minimum effort needed to communicate what he wanted to communicate.

If you are offended by his math gifs and feel that the widely regarded best mathematician of our time should use embedded LaTex or something better, why not offer to upgrade his blog?

lubujackson··on Kimi K3 Is Competitive with Fable; Kimi K3 and Fable Is SoTA
It's a series of tubes
lubujackson··on Agent swarms and the new model economics
That's 100x productivity right there.
lubujackson··on Who Is America's Homer?
Our founding myths are actually Paul Bunyan and Blue. There's a whole cast of figures like Tom Thumb and more, structured around railway construction across America. Sadly, those myths have rapidly fallen from modern awareness, but they at one time were deeply embedded in the American identity, at least in a hokey tall-tales sort of way.
lubujackson··on Who Is America's Homer?
Whitman was first in my mind, he invented free verse poetry after all, embodying the American wilderness in form before content.

If we are going to include musicians, Tracy Chapman indeed seems like a troll when have the significantly more impactful Bob Dylan, who dramatically altered the musical and political discourse at a key time in American history. Tracy Chapman... had a few good hits?

lubujackson··on Infinities, impossibilities, and the man in the white linen suit
This is a fantastic article with lots of interesting questions raised between the halting problem and applications of AI.

The MRI example is very interesting - add a bit of noise and the crisp results fall apart. But we know we can carve out some determinism by orchestrating around LLM (validation loops) but they are only as good as the coverage of validation and the fidelity of expectations.

Which raises a fundamental question: where is the line between an effective harness and a broken one?

To put it in halting problem terms, if we have an outside program that checks the internal program for halting, we know we will capture some number of previously incatchable halting problems. How can we quantify this? We know there is a class of halts we can catch and prevent, but there will always be more complex ones that escape us.

In the real world, knowing the task is impossible doesn't impact the value of reducing the problem. More importantly, knowing which types of halting problems can be fully captured and which ones are intractable. Agent orchestration could directly benefit from layering a framework over this fuzzy space of knowableness.

lubujackson··on Claude Code: Anatomy of a Misfeature
Instead of one-off fixes Claude should have a much richer interface to configure between "ask approval every time" and "YOLO dangerously". I should be able to trivially set "run this task until completed" and have settings like: don't consult the web, don't touch files outside of the codebase, don't delete anything, etc. They don't have to be perfect, just better than the all or nothing system we have now.
lubujackson··on Schema Harness Achieves ~99% on Arc‑AGI‑3 Public
That is a fair assessment, but the "something else" is independently valuable, maybe even more so than model improvements - constructing an architecture for efficient rule determination and execution. In other words, I think the goal here isn't so much to beat Arc-AGI but to develop a generic improvement beyond "Ralph loop", which could dramatically extend frontier capabilities for all kinds of uses.
lubujackson··on Towards a harness that can do anything
I think AI usage is going to fall into exactly this pattern the same way the frontend + API has become the "obvious" approach. We mostly operate with known workflows and just need a touch of LLM magic to determine when to do what.

Right now we mostly YOLO prompts with some docs/skills in the mix but I think it will start to look more like internal MCPs, with tools the LLM can string together. I think the reality will be most tasks end up serviceable by Haiku-level LLM, not Fable.

lubujackson··on I love LLMs, I hate hype
Best comparison is Anthropic/OpenAI are AOL/Prodigy. Massive market capture, no moat. Little by little, the convenience and weight will be scraped off, but they (probably) won't roll over and die for quite a while.

By the same measure, NVDA is Cisco, providing the backbone and capturing a ton of the early benefits, but soon becomes furniture while the excitement moves further up the chain.

lubujackson··on GPT-5.6 Sol Ultra produces proof of the Cycle Double Cover Conjecture [pdf]
Reading the prompt is very interesting. I always wonder how they make these long-running prompts and I guess they literally just tell it to "keep going".

After working with LLMs day-in, day-out an SWE for months, I feel like this could be greatly improved with something like a state machine of progress and proper orchestration. Instead of spinning up a ton of subagents to follow different paths, whip up some Markdown (or LaTex or whatever math-equivalent) to store summaries of attempted paths, and have the agent augment those docs. Leave a paper trail of what has been tried. Iterate on that paper trail and repeatedly examine it for untried alternatives.

LLMs can construct, navigate and summarize exceptionally well. Why is anyone trying to make them "hold the whole thing in your head"? I may be completely off the mark here since I have no math background, but my intuition for how LLMs are able to build on understanding through an external context store makes me feel like this isn't much different than someone trying to one shot a 3D game with Fable Max for $10,000 when they could get the same, or better, result with more human intention.

lubujackson··on GPT-5.6 Sol Ultra produces proof of the Cycle Double Cover Conjecture [pdf]
In my (very) limited use of GPT-5.6, I have noticed it is quite concise in general, and significantly better at abstract thinking. Doing a PR review of a large change it was interesting to see Fable and 5.6 mention a few similar points with Fable much more long-winded and less readable, while 5.6 caught more "second-level" concerns and Fable more "in the code" concerns, so they both are quite useful in concert.

In general, I would not be surprised if 5.6 was a much better tool for high mathematics than Fable based on the abstract thinking. For my dev workflow, I have flipped my approach from planning with Opus 4.8 high and implementation with GPT 5.5 to planning with 5.6 high and implementation with Fable medium (and I might even drop to Fable low). This is only on the company dime, of course.

← PreviousPage 5 of 34Next →