softwaredoug

I wonder how long it'll pay for frontier labs to use compute this way?

I'm reminded of the Apollo program that was cool for a while, but eventually the public lost interest.

Maybe it becomes cheaper? Maybe other labs want to prove their worth will pick up the slack when another loses interest?

It’s not the software development that’s hard. It’s the data science and domain knowledge to build a good training set and a meaningful objective.

This is like saying we don’t need Dropbox because we have FTP servers.

People vastly underrate the patience of people to learn and setup a classifier. Jev is one API call. Training a good classifier is hard.

I discussed taking a PM role at a mid sized company recently. They were so locked into their classic definition of PM, and so against considering with AI a true PM must straddle product and engineering competencies. I felt a bit sad for their org, and I turned down the role.

We’re not ambitious because the economics are unimaginative cost saving. Large, complex orgs turning a team from 10 developers to two is the easy optimization. Meanwhile smaller companies are more focused on discarding old role definitions that don’t make much sense, and embracing fluid roles mixing tech and business.

Yes exactly. And I’m not sure we want agents with that level of autonomy to decide what functionality is important and what to leave in place.

IMO slop usually comes from overly defensive programming from LLMs. They’re loath to remove code and simplify, because unless we humans tell them some code / use case is unimportant, they have no reason to not care about it.

So agents can make these overly complex solutions trying to make bulletproof implementations and prevent regressions.

It requires a human to look at code, ask about complexity, and guide simplification by subtracting requirements implicit in the code.

The real solution to this has always been education. Teach the kids about social media the way we teach about drugs and sex. It’s in many ways more ubiquitous to many teens than the other two these days.

(Oh and regulate the companies, not what consumers are allowed to do)

No this isn’t about hacking voting machines, it’s about controversial defense work to stop outside foreign influence on social media.

Interesting this isn’t prompting Claude with “build me a version control system” but someone hand wrote version control approach, then worked with an agent to improve / reimplement the idea. That matches my notion of the role of writing code in 2026, to learn and iterate on an approach. Eventually an agent night rewrite or improve it.

At this price I’d also want a Dolby vision Blu-ray player. Strong overlap with this product and physical media people.

It’s highly structured in a way my teenage years were not.

IMO a lot of kids now experience their adolescence in college, not in high school. High school kids focus so much on activities and schoolwork, they don’t get the same social experiences they used to in high school. Now they get it in college.

College becomes the safe island for young adults to grow into their own, gain social skills, and grow relationships for later in life (professional and otherwise).

Which also is interesting because what does this mean for non college bound people?

I stand corrected, I'm glad to hear that.

EDIT - glad to hear that Character.ai employees did not get screwed, at least per a comment below

---

I worked at a company in 2000s that was acquired. I had no stock but the owner still gave a generous spot bonus and a years salary bonus retention package to employees. Owners got life changing wealth, employees got a down payment for a home.

It’s sad that this doesn’t seem to be the norm. Screwing employees on acquisition seems like a severe breach of social contract.

I think of the point of vibe coding is I make my own version. I don’t expect or want other people to use it.

I built my own Jekyll markdown editor. Supporting anyone else using it would be annoying.

These articles IMO suffer from good data. It seems we could probably at this point study open source projects longitudinally and see what we learn depending on their AI policies. Are there patterns in stability? Bugginess? Performance? Readability? Etc.

One thing - your expenses will ALWAYS somewhow expand to your income. We see our current income as somehow the steady state long term norm. I would do your absolute best to hide extra income from your day to day self and invisibly invest it in a way you don’t even think about

And when you get a windfall (stock grants, acquisitions, etc) unless it’s truly life changing money where you never have to work again, find a good investor (CFP) to manage so you’re not even thinking about it. Or invest it in some future liability like college expenses, home down payment, retirement, etc.

Leaving this stuff at the discretion of courts will mean a lot of routine filming of police will just get thrown out by right leaning courts.

No because then courts would litigate what “newsworthy” means leaving your 1st amendment rights in the hands of random judges. I can imagine filming ICE doing routine operations being seen as “not newsworthy” by a conservative court.

All I’ll say is… see Digger before it leaves theaters

To me the Jev moment isn't about Jev / Typesafe. It's articles like these. It's the letting a million Jevs bloom.

The bubble will come from a realization that a lot of people can train these models. AI Researchers will become more diffuse, work at more companies, and building a Jev or fine-tuning an LLM isn't some trillion dollar frontier lab exercise, but increasingly just something developers do.

I’ve started to see (and react negatively) to a kind of “Claude design”. Apps, slides, images share the same aesthetic, and even after aggressive prompting it reverts to a the mean.

I even create these things. But that aesthetic has started to imply a kind of throwaway, low value thing

What will be interesting is if models can create more diverse styles than the neutral Claude default we see everywhere

Not the optional war Trump started where 1/4 of the worlds oil is shipped

Why is it a middling product? Most clones don’t approach its performance, and it solves a specific problem well in a way that was awkward and ignored by most frontier labs.

What’s interesting about the Jev moment isn’t just Jev, it’s the unleashing of distillation / fine tuning outside the frontier-adjacent labs. It’s the sudden explosion of a million Jevs.

If being an “AI Researcher” is a ticket to multimillion dollar salary, AI training talent cannot be contained to a handful of companies. It’ll become more common and diffuse. The old advice of not fine tuning, because it’s hard, goes out the window as that knowledge diffuses through the industry.

A similar thing is happening in search. For a long time labs have trained tailored embedding models. And now companies like SID training their own agentic models that are smaller and faster at search than GPT-5.

Probably the same reason people aren’t vibe coding nuclear reactor software. One small mistake has large downside consequences.

To be frank, the obsession with being first, and not making research accessible, has always held academia back

Aren’t there dozens of proofs of the Pythagorean theorem? The goal isn’t to just “prove” but create something well written and intuitive to the average practitioner. And by gaining a deeper understanding we can ask better questions.

Clearly AI can grind on math problems now. It can generate proofs and get immediate feedback.

But what I wonder: can we legitimately grind on Physics or curing cancer? There’s a lot of physical world experimentation that needs to happen to make progress.

The question to me isn’t whether they can rush out an API, but whether a general model can out compete one post trained on the classification task. GPT-6-Luna has to write emails and classify them. Jev only needs to do decisions.

Or does OpenAI start distilling their own models for use cases like decisions?