Something I hate about ChatGPT is that it assumes I want my text to be rewritten instead of engaging with the content.
I like my writing style. Sure it may leave some sort of linguistic fingerprint and it may not meet some LLM’s idea of what “good” looks like, but I don’t care.
What’s worrying is that the rewrite-by-default behavior is probably there because most users want it.
I can't allow LLMs write my messages in any language I'm fluent in. They just don't sound like me at all. It would feel dirty and dishonest to send that slop out.
They can be good at grammar checks, but even then I wouldn't fix quite everything, it's better to let some of my natural flaws go through.
My wife translated some text she had written in her own language to my native language, as she didn't feel she could do it.
I found grammar errors, surprising ones. I found places where pronouns were used so far into the sentence that a reader would be lost finding what that pronoun actually referred to. And more.
In the end I rewrote the LLM-generated text myself, many pages of it.
I know this is going to. be contentious, but US mainstream discourse seems to have completely eliminated the distinction between illegal and legal immigration, in the last 10 years. Everyone seems to be a "migrant".
US policy has also nearly completely eliminated the distinction, by making legal immigration close to impossible and ~arresting~ kidnapping people at courthouses who are there for their immigration hearings, then shipping them off to foreign torture camps.
It is so nearly impossible, that somewhere between a half million and a million people have done it every year for the past few decades (including last year).
I don't think that is true at all. For example, it was considered a big deal when ICE was rounding up US citizens. It caused a big drop in public trust for ICE.
Nearly half of the workforce of crop farmworkers in the US is made up of "illegal" immigrants. The US food-supply relying on those people has meant that, in practice, immigration law enforcement is deliberately selective and self-serving.
So, the idea of illegal immigration as a vice worth cracking down on and punishing has not been consistently applied by the people publicly condemning it (like this current administration), meaning there is a very real sense in which the distinction between illegal and legal immigration is not real.
"The people"? Are you sure you're not committing the common sin of conflating vocal people with most people? For example, I publicly condemn illegal immigration, regardless of which industry said immigrants are propping up, while at the same time recognizing that such industries need to be carefully extricated from reliance on illegal immigrants and also that the management of immigration and the definition of illegal immigration needs to be fixed.
Tesla’s Solar Roof uses string inverters rather than micro-inverters or power optimizers, which means that partial shading on any section of the roof can shut down production for that entire string. This is a significant design limitation that competing solar installers address with panel-level optimization technology from companies like Enphase and SolarEdge.
This seems to be overblown. I've seen plenty of string inverters around without issues, I'm not sure why this being used against Tesla in particular.
The clean linear history thing is something I never really got, despite using git for 12 years now. I worked with some smart developers whose rule was "rebase if you want, but if too complicated, just merge", and it didn't hurt the delivery or maintainability of the code they wrote.
Yes - whenever I'm in a team and I hear someone who insists on a linear history, I always wonder why they have trouble with merge when lots of folks like me have no problem with it.
Finally, in one team, I more or less forced a senior engineer use merge (or rather, I was in control of the project and did not force other developers to use rebase). After a year, he admitted that he no longer really saw a benefit in rebase and switched to just using merges in his own projects. He also noticed fewer merge conflicts this way.
Rebase makes sense when you realize git doesn't have branches. Git has tags that move but no branches. That means when you merge you have no clue which branch was the mainline and which was the fork. This is a question I often ask 10 years after switching to git. Sadly git has better tooling so it is worth using despite the issues.
> That means when you merge you have no clue which branch was the mainline and which was the fork.
You mean - when looking at the history?
Incidentally, once you get used to jujutsu, you realize that the question is meaningless. A merge is simply the child of two nodes. It's a symmetric operation between the two branches. The thing that makes it "complicated" in git and traditional VCS's is the insistence in assigning a name to the resulting merge (so if you're merging into main, you want to call the new node "main"). Since jujutsu doesn't automatically carry the name forward, you see the "reality" of merge being a symmetric operation (i.e. you don't merge a branch "into" another branch - you are simply merging two branches).
That is exactly my point. I'm not merging two branches together. I'm merging two branches with very clear different meanings together. One of them is our main line, one of them is a feature branch. Everyone talks about all you should develop in main line, and I certainly encourage that. However, often that just isn't practical in a large project for various reasons. Some of them aren't even good reasons, but nonetheless that is the reality.
For a complicated long running feature branch I can see it. Instead of repeatedly merging the root in during development it can be cleaner. Tools aren’t always good at figuring out in a PR what was written and what was caused by those merges from root. And history looks better at the end.
For a short branch that can merge cleanly or perhaps very close to it, I’d kind of rather have the ‘true’ history. I don’t think it’s worth it.
I’ve never understood the “everything must be rebased before every merge” desire.
It really depends on how often you use git bisect and blame. This varies greatly across projects.
That said, if/when stacked PRs become a first-class citizen in GitHub, more projects will see the benefit of this approach (though they'll probably mostly get there through squash-merges).
Engineering has always been about more than writing code.
That's true, but it's interesting how FizzBuzz as said to be the bete noir of the average dimwitted software developer, and how much cutting-edge engineering organizations used to emphasize code in their recruitment processes.
If writing code is being replaced by "engineering judgement" it's going to need a much smaller cohort of developers. Too many opinions spoil the broth, after all.
I suspect that the goalposts for AI-assisted coding will be moved the same way they've been moved for the Turing Test.
The Turing Test used to matter until it didn't (does anyone even talk about it? was there a big news conference when it was solved?). Likewise every time it becomes easier to ship software, the bar will be pushed higher by sceptics. Ultimately the gatekeeping is going to become meaningless as software becomes "too cheap to meter".
For something that was supposedly always unimportant, huge amounts of energy were spent recruiting developers based on how they produced and interacted with code.
FizzBuzz was a litmus test that showed how hopeless the average developer was. Coding interviews were the real test of programming ability. Now we're being told none of that ever mattered for real?
We should just admit that the game has changed (possibly, I'm not 100% convinced). Code WAS the bottleneck and coding ability was the bottleneck, but it may not be going forward.
This is clearly a well-timed loss-leading strategic market share grab! Anthropic have blown a lot of user trust in the last couple of months..
But, overall, the current AI pricing is completely unsustainable, across all AI companies, except via the exponential growth they are relying on. Dylan Patel did the most insightful analysis of this I've come across.. https://youtu.be/mDG_Hx3BSUE?si=nyJu4adwYCH1igbJ
Really feel like the current versions are for sure "good enough". Thats not how market capture is gonna function though and they are gonna keep pushing because the only moat is to stay ahead, so the problems gonna stay strange. at some point more compute isn't a reasonable answer, and optimization is, and my feeling is we are well past that point from a product perspective, but ipos etc etc
The only moat is the us trying to buy all the compute hardware in the world for the next two years. Then China, amd, etc are just making their own chips.
So I think the current generation of models are arguably all about the same in terms of capability. However, the requirement for exponential growth I mentioned is all about the economics.
AI companies are trying to ride a growth wave where the income curve lags the expense curve by 1-2 years, and at the same time investing 10x their historical income on next year's projected demand.
Everyone is selling their API calls at a loss, because to capture the investment required to scale the business up and the costs down, you need to grow your market now (in relative and absolute terms). And history shows, that in big tech you often have winner-takes-all situations, or, at least a couple of big firms will dominate, and the others will die. That's where market share becomes a key strategic goal.
But to secure that, they also need to be building next year's compute now. And if their anticipated compute needs are 10x this year, they've got a serious funding problem, one that can only be filled by capital with an appropriate risk appetite. You can only get this high-risk capital when the potential payoff is even more enormous, or, when it's a smaller bite of a much bigger pie. Hence, MS putting into OpenAI and so on. But the investment needs are getting so big we are starting to see some pullback from more conservative sources, but also record deals from others.
Now say an AI company does get the capital they need to grow. Well, they've still got a very serious supply problem. RAM, GPUs, water, electricity etc. Hence why there's a lot of deals and cross-investment going on - everyone is trying to secure resources and lower their overall risk exposure while keeping a foot in every possible door, so they can switch alliances whenever it's expedient, and because collaboration also helps the overall market to grow.
This all explains to me why the industry _needs_ the hype. These companies can't exist without it, because the money they need to sink in, in order to even be around in 18 months, far outstrips all reasonable financial practices. So it's capitalism on steroids or nothing. If you believe the AI story, then to that extent, it's rational.
But note that nowhere in this scenario does it suggest the actual consumers will be getting a consistent product at a consistent price!!!
Qwen 3.6 supports reasonable agentic programming. People are vibe coding with it. It's really not that far off. If you truly cannot make a model that was SOTA 6-12 months ago work for you today for free I don't want to know what your needs are.
Free lunch? More like "free data". The fools who give their life data and most intimate Intellectual property over to the AI companies for free, yes that's a free lunch that won't be subsidized for much longer when the cost on them which has been unsustainable (their data being harvested for non-training purposes) come stop catch up with them.
Sincerely,
- I see you AI companies harvesting our data giving us discounted subscriptions so we can not realize we are paying you to take our own data!
They need to build data centers and lots of them everywhere, preferably powered with renewable energy. Let the tokens flow like water. The models are finally getting to the point where the LLM just knows what you’re asking for and gives it to you.
there will be free lunch till they admit to themselves that there is no moat. Acquring customers at huge costs is a fools errand when models are mostly indisguishable.
Anthropic is learning that lesson now. Doesnt help that their ceo goes around antognozing everyone by claiming jobs are over and annoying boris does like 500 podcasts per week repeating "coding is solved"
I’m not happy with their privacy policy [1]. I’m unfamiliar with the phrase “Parties with Other Legal Rights”. Given the well-documented struggles of Anthropic and others to provide enough compute, I wonder if “Parties with Other Legal Rights” constitutes part of the advantage here.
Just run a local model or run deepseek from another provider with a policy you like. The models are open weight and widely available. Still cheaper than chatgpt and anything else through 3rd parties
this is the pitch - it's open source, run it yourself. But >99% of people will not have the hardware needed to run these models at a high enough quality to be close to SOTA. So they will run the open-source models on CCP systems for a good price.
a lack of existential threat in the form of pay-seeking and remediation from the people you stole training materials from that allows for an intrinsically different pace of operation than the Western competition
I can't figure out how there's both too little supply (so a dramatic need for more data centers) but also too little demand (so labs subsidize inference).
There isn't too little demand. There is massive demand and many competing companies trying to capture that demand, so they are attempting to make better offers than their competition. Hence subsidy.
- Every competitor is planning for the demand to be much higher in a few years than it is now, and aiming to capture as much of that as they can, which starts by getting companies hooked on their models now
- The data center capacity will get used no matter who captures the most demand
I can somewhat understand companies getting users depentant on their harnesses or workflow, but model vendors as in this deepseek case, I have absolutely 0 model loyalty when it's a simple config change away, and will always optimize for either capability or price (or whatever !/$ metric you can determine).
Depends what you’re doing. For example, Gemini is somehow still your only option if you need a model that can natively understand video and reference timestamps in its response.
In my experience, managers don't have to be hands-on, but they need to be able to recognize people with talent and unblock them do their jobs, to be able to spot process improvements, including channelling the AI hype to productive outcomes, and to be a steadying influence in a crisis (without adding noise). If a manager doesn't have technical ability, its impossible for them to do those things.
Everything but the AI bit are on my list of manager qualities too, but the best managers I've had weren't active programmers, and one had zero coding background.
Knowing what you don't know and knowing how to get qualified information from people around you makes up for a lot of not having a programming background.
If anything, the managers with technical backgrounds who weren't active programmers tended to significantly underestimate the difficulty of doing something because back in their day, things were different or some such nonsense.
I think they simply just haven't figured out that the barrier to entry is so low, that no one really cares what their app can do, even if does something genuinely useful.
I like my writing style. Sure it may leave some sort of linguistic fingerprint and it may not meet some LLM’s idea of what “good” looks like, but I don’t care.
What’s worrying is that the rewrite-by-default behavior is probably there because most users want it.
reply