I've jumped between copilot, claude, gemini and chatgpt since the start of the year. chatgpt wasn't even worth looking at early this year.
Anthropic has the smarter models for sure, and seems to be default in corporate. However, the amount of budget you get with GPT as a user is much better, the harness feels more polished, and the models are faster. They are also much nicer to work with, I can just read the output for the most part. With claude I get pages of text and need to skim to find where the actual information i need to care about lies. So much more cognitive overhead.
Sol is smart enough for anything I've thrown at it, it's not one-shotting like fable, but I'm more willing to actually go back and forth with it, and it's likely producing better output to keep a human in the loop rather than trying to solve the world independently and making multiple incorrect assumptions.
I think GPT sees the market changing and is correctly repositioning themselves. Anthropic is down the wrong road, and if they don't correct course quickly I'm sure many of those enterprise contracts will start pivoting.
I was just discussing this with a co-worker yesterday. I would really like a model (or harness?) that worked with me instead of for me. Walk me through its choices and decisions, let me correct it and guide it along. I would be way more confident in it's output, I would be more familiar with the changes that are being made, and it would make reviewing the final code way easier since I was making the decisions along side it. I'm sure it would also reduce the "brainrot" we're all going to experience the more we hand work to these models.
I'm not sure how they'll survive their creditors tbh
The collapse of the massive overvaluation and circular investments will be the undoing of several major tech investors and companies that have a serious contribution to the things in tech we don't like. It will be bad for the economy, but it will also weaken the abhorrent control that those companies have over regulation in the United States.
It's gonna be a rough time, but it's pretty clearly necessary.
The only big problem is that if that collapse happens under the current administration in the US, they will be utterly incompetent to respond to a real domestic crisis, since they have been virtually unable to do anything without creating more problems.
They're both pretty damn competent and more important, Extremely Fast! I find the speed more useful than trying to be a hundred percent complete on every task. The big intelligent models screw up all the time as well, but I have to wait twenty minutes to three hours to find out.
GPT 5.6 is especially tenacious and seems to want to solve every bug in edge case 1000% all the time. Sometimes that's what you need, but a lot of times you're just trying to move fast and figure out what the product is.
My boss uses opus and gets good results when I use it always burns tokens. The other way is also true too I get great results with sol/terra but my boss does not.
This has been a recurring problem for me - as a security engineer catasrophization is a fundamental skill to finding vulnerabilities in complex systems, but makes me to conservative when building. For some of my leaders they are much better at saying 'good enough, ship it, and fix it later'.
“No, not like THAT!”
When I checked, all that credit was gone, I still wasn't into the next 5 hours, and all the agents had failed, returning nothing.
I didn't even get anything for burning all that credit. If I had paid for it, I'd be very, very pissed.
One fine day I was like, oh my weekly quota resets soon, I will kick off an expensive bug hunt. It launched parallel things and burned $10 in like a moment.
It’s scary. I turned off extra usage after that.
"Ultracode. When a system-reminder confirms ultracode is on, that opt-in is standing: author and run a workflow for every substantive task by default. The goal is the most exhaustive, correct answer you can produce — token cost is not a constraint."
gpt 5.5/5.6 goes further on its own much more often than opus 4.8/5 does. codex capped ~300k context when claude does 1m.
I don't feel codex is saving tokens, and result is usually not as good imo.
That's configurable in codex.. but there is a higher cost/usage to using it.
By contrast, Claude Code's bias to make assumptions of reasonableness about underlying systems has proven to be immensely frustrating over the last month or two, both personally and at work. I've wasted days on "that was my mistake. I've been reporting numbers on the old architecture because I hadn't enabled the new one in the config" both at work and home. It's immensely frustrating.
But here we are. Wrestling with energetic idiots in model form, wrangled by over-specific harnesses that struggle to stay off of deranged side-quests.
What a time to be alive!
Granted it did make me think about my biases and to lean into more future facing inevitabilities. But you've nailed it, for Boris and Anthropic, they are betting that coding is a solved problem in the sense that any person can one-shot any random idea and the output will be in some abstract sense "good". And then at what cost and toward what end?
As someone who does near 100% of my coding via LLM these days, i still find that for anything complex i am still looking at and thinking in terms of code. Im still quality checking and steering at some interval via code. And im still not sure how or whether i can replicate that level of thought without still dealing in code at times.
Isn’t the race between Chinese open-weight models and the others more decisive for the future?
once we have a bit more memory fab capacity and the insane bottomless investment in AI giants realizes there is a bottom, local hardware will catch up with model performance to the extent that centralized inference will be downgraded to special cases or for orgs that find it cheaper than buying expensive hardware
but really most power users are going to have 1 TB unified RAM and local models that will do well enough
for light office use you can still have a cheap laptop and a claude subscription
I find it unlikely that there's some fundamental property of OpenAI's models' "personality" or style which Anthropic (or any other serious AI firm) wouldn't be able to match if they wanted to.
“In one notable technique, their prompts asked Claude to imagine and articulate the internal reasoning behind a completed response and write it out step by step—effectively generating chain-of-thought training data at scale. We also observed tasks in which Claude was used to generate censorship-safe alternatives to politically sensitive queries like questions about dissidents, party leaders, or authoritarianism, likely in order to train DeepSeek’s own models to steer conversations away from censored topics. By examining request metadata, we were able to trace these accounts to specific researchers at the lab.”
— “You are an expert data analyst combining statistical rigor with deep domain knowledge. Your goal is to deliver data-driven insights — not summaries or visualizations — grounded in real data and supported by complete and transparent reasoning.” (variations appearing 10s of 1000s of times)
https://www.anthropic.com/news/detecting-and-preventing-dist...Does Dario have the same relationship with the truth as Sam? (Their companies pirate books and develop products that compete at some level with those books’ authors, so obviously neither are that wonderfully trustworthy, so maybe “no evidence” meant you don’t believe this evidence rather than you weren’t aware of it. I would understand and respect your lack of belief!)
You break you pay .
That’s why in the western world you got book publishers . They have lawyers etc .If you prefer Chinese way : laugh at anyone who is taking about copyright , whether that a book , a model , or manufacturing secrets in eu , us or whatever sure .
A family business of my friend in eu was destroyed due to stealing until every single bolt position design of their system by China . They had no legal protection .
Keep calling rule of law vs no such rules an American exceptionalism .
It’s economically lucrative for everyone in that chain to attack again and again ai vendors , unless they are in China . In this case they can only cry .
This is evidence only if you're an American exceptionalist.
“These labs generated over 16 million exchanges with Claude through approximately 24,000 fraudulent accounts…”
and think “oh, there’s alleged evidence of distillation“.https://en.wikipedia.org/wiki/American_exceptionalism
Seymour Martin Lipset, a prominent American political scientist and sociologist, argues that the United States is exceptional in that it started from a revolutionary event. He therefore traces the origins of American exceptionalism to the American Revolution, from which the U.S. emerged as "the first new nation" with a distinct ideology, and having a unique mission to transform the world. This ideology, which Lipset calls "Americanism" but is often also referred to as "American exceptionalism", is based on liberty, individualism, republicanism, democracy, meritocracy, and laissez-faire economics; these principles are sometimes collectively referred to as "American exceptionalism".
As a term in political science, American exceptionalism refers to the United States' status as a global outlier both in good and bad ways. Critics of the concept say that the idea of American exceptionalism suggests that the U.S. is better than other countries, has a superior culture, or has a unique mission to transform the planet and its inhabitants.
Like, when I consider the Tuskegee Syphilis Experiments or MK Ultra, or Flock, I have bad things to say about them but American exceptionalism specifically wouldn’t cross my mind. Corporations anywhere playing a card close to the chest or even lying… superiority, exceptionalism, aren’t my top reference points.everyone else uses it as "Americans love to criticize everyone else, while pretending their farts dont stink"
Perhaps partially covered in the “Critics of the concept say…”
Certainly weird to think the US isn’t chock full of problems (even if historically some stuff has been neat, like bringing together cultures and developing nifty tech in Silicon Valley).
As much as I wish you were right, everything about software economics for the last 30+ years has favored _less efficient software_. Traditional hardware has been optimized for traditional software for decades and we still see bloated software win consistently. LLM hardware is at the start of its cycle, with abundant low-hanging fruit to conquer--I would expect the pro-bloat dynamics to weigh even more heavily in the LLM space than in the traditional software space.
I'm just happy there is capital behind the efficiency path because I am confident we can get positive value on that path.
Open weights, obviously beneficial. Local compute when not necessary, seems like it'd be significantly worse for the environment?
I think because more devs cant run things locally, we are not getting the "open source gains" that you would get by having the network effect of more eyes on the work. Theres still little nuggets of gold in there though (eg antirez's DS4 project). But once more people can get their hands on hardware, I think the flood gates will open. And once that happens, I dont necessarily think people will run locally. But that will enable fungible service providers which we already see, but at a much larger scale.
To a lesser degree, open weights also leads to an eventuality of these two companies not being able to sustain their capital investments. Not to mention their pricing models are unsustainable as they currently stand. So they are being eroded from the outside-in, while also trying to fight the internal pressure to continue to raise prices/margins.
EDIT: I actually think the harness is more of a future for these big players. And theres real value in providing a high quality harness. So maybe they move to SAAS for claude code in the future. Who knows
My money is on the more efficient solution. Even if Anthropic can win some benchmarks by using 3x tokens over 3x time, it is a terrible base to build toward the future. Users are no longer willing to wait exponentially long for linear improvements. And as we see from some of the Chinese models, they can quickly distill frontier models with the tax of being slower and more token-guzzling, while retaining most of the quality. The real differentiators are becoming speed and efficiency, which translate to cost and user velocity more than incremental capability improvements.
It's great that frontier models can solve complex math equations, but bread and butter LLM usage (where the money is made) has already shifted from "I need the best always" to "what solves my day-to-day problems quickly and consistently". Fable usage as a percentage is flat-lining. We are already at the point where output quality is negligible. What wins going forward is cost, speed, consistency and the compounding effects of "softer" improvements to the harness.