lubujackson

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.

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.

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?

It's a series of tubes

That's 100x productivity right there.

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.

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?

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.

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.

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.

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.

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.

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.

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.

Even simpler, use Cursor with any frontier model. I see others sweat to add enough context to Claude Code while Cursor has a ton of contextual awareness, uses subagents automatically and is significantly faster with no drop off I have found. I'm not sure why devs are so enamored with living in the CLI, but Cursor has one of those too.

I think the reality is that you should focus on and serve the 2c group (best students who use AI to fill gaps or speed up learning). We have 10s of thousands (or more) students taking advanced math every year. Only a slice of those pursue PhDs and even fewer contribute anything meaningful.

This is how it has been for years and AI doesn't really stunt that as long as students actually care about learning the material. Those kids will be fine. The kids that are fulfilling a prereq will be fine, for the most part, but will walk on crutches their whole lives (use AI) which might be a reasonable choice.

What we lose is the signal of which kids are interested in learning and which kids aren't. Grades and effort used to be the signal, now I think it has to be more about engagement" who asks questions, who comes in with strange ideas, who actually is struggling at times (actually grappling with the material) or learning things outside of the assignments.

Big picture, I think teachers should focus on teaching the learners and trying to motivate more students to join them. That has been the real job all along, AI just lays it bare.

You need to know what to ask and what the answer means. Also, using proper terminology greatly increases LLM accuracy to situate the response in the correct domain.

Ask an LLM to "clean up this code applying SOLID principles" is much more prescriptive than "clean up this code".

This won't happen for the same reason very few professors stand up to this shit and why everyone gets straight A's at Harvard now - Goodhart's Law again. The money involved means getting accepted into the school is the whole task, grades are irrelevant and learning is opt-in at best. Stamp the kids, extract the money, funnel it up to the administration.

Rape vans, considering gas prices

This question, and many "but make it a bit more challenging!" comments always strike me as CS101 navelgazing type questions. The best part of this question is that it is simple and can be used to swing into deeper concerns but it is still at odds with actual job responsibilities - even more so with LLMs in the mix.

Maybe this is because I have only worked at startups, but I am much more interested in if someone can read and understand code, where they feel logic is brittle, overly complex or badly designed. If they understand, even conceptually, how adding an optional field to an endpoint may be fine but removing one needs to be phased out or considered for active users. If they consider downstream risks, if they understand business goals and how to communicate limitations or opportunities.

Instead, every single tech interview seems to focus on how well you paid attention in your CS seminar which might be a reasonable screen for junior employees but is awfully irrelevant for anyone >3 years in the industry.

There are far too many corners of logic for everyone to know. Maybe someone has never dealt with data streams, or even forgets what a median is. You want to know if they are sharp with statistics? Great for some roles, wholly irrelevant for many others.

Engineers need to communicate, read and understand logic and how things connect. And the golden skill: willing and able to learn something new.

That assumes the shift was entirely due to high school students becoming adults. There are also people who haven't read much over those years and have had their skills declined to the point of failure. Or, also likely, sample size issues.

LLMs are just a dumb terminal related to permissions. What they apparently want is some synthentic permissions spun up based on their prompt which is... not a "prepared statement" solution and more of a "I will clean user SQL statements with my handy regex" and we know how that works out.

The real solution is a better UI for controlling permissions on a per prompt basis - just as we can select "search the web or not" the solution would be to have a "include my private repo" option that can be trivially toggled.

Makes sense, I have had the biggest wins with AI by attacking nondeterminism whenever possible.

BTW, you should probably fix the Beagle link on your homepage: https://replicated.live/beagle/

Whenever I see keyboard-mashed company names I know I can go to Temu/AliExpress and get it directly there. You can tell when they are all sourcing from the same batch and instead of paying $40 the same thing is often $3.28 if you don't mind waiting a week.

Yup. Once you realize neither party actually cares about any of the top debated political issues (individuals do, but the party doesn't) but they thrust them in everyone's face day in and day out BEACUSE the country is divided neatly in half and they are emotionally charged: abortions, LGBT rights, gun control.

Let the plebs wear themselves out so money can be extracted and power can be used without any fuss.

Very cool tool. As the "moar tokens" era is starting to wind down I think people are going to realize just how crappy these harnesses really are, especially Claude Code.

I have gone back and forth between Claude and Cursor and it is clear Claude just throws the kitchen sink at problems to get an edge. I write MCP tools and I see these exact problems when the inputs and outputs aren't clearly defined, the LLM just guesses and retries.

The point is, what people are interested in is more social than inherent in the thing itself. Art is communicative and socially-bound.

I have gotten a lot of mileage out of giving LLMs narrow focus over the same plan or code change with different objectives. Write a plan, ask the agent to review the plan and consider where code can be consolidated. Consider downstream effects. Consider security risks, consider optimization, consider architectural concerns, etc. Then I generate the code and go through a similar loop. Then I read code and do more loops to clear out any problems or investigate things I'm unclear about.

In this way I spend most of my time building understanding of existing code and understanding the impact of my changes. My company is heavy into AI use and I find I am pushing out more code and much cleaner code than most. The gaps that appear during review are usually product understanding gaps and not code failures, and my LLM spend is somehow less then most.

I find this iterative process is much more inline with building flow than spending 3 hours writing a spec and wait a half hour for it to build a monolith PR.

I definitely agree with the core concept here - I think the the future isn't "make it so" prompts but crafting architecture and only sprinkling in nondeterminism where necessary. If a process can benefit from a DAG, it has no reason to give control to an LLM, so finding where to make the handoff is key.

I will say this has the stink of going too deep into the AI brain. It supports everything and can do everything and has a million options, but it lacks specificity and purpose. Humans need a simple "how does this help me immediately" to get excited about trying something out, not a laundry list of features when the use case isn't obvious.

So fans have fan fiction because they love certain worlds and characters and want more of it, or to riff on the base to take stories in new directions. We have things like TV tropes so it is no small stretch to, say, generate sitcom-type stories from Harry Potter or GoT that can continue forever. But where does all of this lead?

Music has sinilar issues, but I think of this like the compression/loudness issue in music production. Everything gets amplified so the range of everything is compressed. And then it gets boring and people slowly jump ship for something else.

I think there will be a wave of AI slop that improves and might actually be exciting on some vector, but we will get bored eventually. Humans crave newness, even if that new thing is worse in exactly the ways that defined good previously, like punk in response to complex rock albums.

AI can combine ideas in interesting ways, but it is by design a predictor of what is most likely. This is directly odds with the concept of newness (and arguably, human-ness) which is baked in to what we consider interesting and relevant.