thimabi

What do you suggest for articulating the writing style that one wants from LLMs?

I’ve been experimenting with having LLMs write/update academic notebooks for me, and so far the best results I’ve gotten came from correcting their output and asking them what they’ve “learned” from my feedback.

I noticed the same thing. In the app, normal “Chat” threads are only available via a “Checking recent chats” window, while projects, GPTs, library… are completely absent.

The web version of ChatGPT is confusing too. Now it has separate “Chat” and “Work” tabs (what about a Codex tab?), and it shifts the burden on the user to know when to use one or the other. Note that using the “Work” tab means using Codex usage limits [^1], but that’s hidden away in the settings.

Also, apparently “GPT-5.6 Terra and GPT-5.6 Luna are not selectable in standard ChatGPT conversations” [^2] — so if you want to use these models, you must go to the Work tab or download the app.

I don’t understand why there is so much fragmentation in what was supposed to be a unified app. The way that it is now, it’s far from intuitive.

[^1]: https://help.openai.com/en/articles/20001275-chatgpt-work-an...

[^2]: https://help.openai.com/en/articles/20001354-gpt-56-in-chatg...

I’m interested in knowing how each of GPT 5.6’s variants fare in non-English writing/translation tasks.

GPT 5.5 has a tendency to write English calques and non-idiomatic prose in other languages. Although that can be somewhat tamed with detailed instructions and a corpus of confusing terms, the model’s output often reads like a literal translation rather than native prose. Since I notice these issues most clearly in languages I know well, it makes me reluctant to trust the model’s output in languages in which I’m less proficient.

Ironically, ChatGPT began as a simple text-generation tool, but much of its offerings and benchmarks now focus on coding and agentic workflows, while leaving behind what made it notable in the first place.

I wonder why they haven’t simply continued to rebrand Codex as a general-purpose tool. ChatGPT Work is a convoluted name and continues the trend of having separate brands for separate things, what runs counter to OpenAI’s purported goal of unifying every workflow into a single “superapp”.

Worse still: what happens when your workflow involves both coding and general knowledge work? Are you expected to switch apps, or switch settings? To me, it sounds very confusing and inefficient, and not at all what I was expecting.

In a few years Sony executives will be wondering why a portion of their consumer base decided to prioritize other forms of entertainment. I can speak for myself in that I’ve never upgraded past the PS3, and I feel no regrets about it.

I’ve been experiencing similar feelings. Working with LLMs often takes almost as much mental energy as working with people, but the payoff does not always scale in the same way.

I think we are still on the early days of LLMs. Right now, using them productively requires deliberate thought and an acute knowledge of their limitations. As the author says, it’s easy to get angry at a model, or to foolishly let it nudge you towards more code and more tests — even when that is suboptimal.

To a certain extent, models keep getting better and better at discerning our intentions and providing value. Yet I am not sure whether we will reach a point where using them successfully no longer causes the kind of fatigue that it does today.

Buying a camera or car is different from paying a subscription, right? Different expectations

Surely there should be a way to enable deep links with a fallback to Safari without needing Gmail to know what apps I’ve got installed

I got 68,900 words, with the vast majority of the errors being on the grandmaster level.

As a non-native English speaker, I found that result pretty good! Though being a native Portuguese speaker certainly helped me as many difficult words in English borrow from Latin, and in Portuguese the Latin influence is more pronounced.

Yak shaving with AI allows me to function more as a systems designer, code reviewer and tester than a coder per se.

AI is great if you simultaneously guide it and let it guide you. I take my time building a very detailed spec for what I want, then run it through the AI looking for contradictions, misconceptions, edge cases, performance bottlenecks, potential optimizations… anything that might cause problems in the future. Usually these discussions lead to multiple spec-improvement journeys, and that’s where the bulk of learning in a project comes from. Sometimes the AI will flag actual issues, while other times I might need to rein in its proposals — mostly in terms of feature creep and finding non-existing problems. I believe this back-and-forth is the most significant aspect of making the best out of yak shaving.

By the time the spec is “final”, it can be quickly implemented by an AI as I watch, review and test, with practically zero code banging on my part. This way, I get to understand precisely how the project works, make it tailored to my needs, and still not waste time, muscles or even mental bandwidth with menial coding.

I always liked yak shaving, but avoided it because I knew it came with costs and tradeoffs. More recently, with the help of AI, I’ve been doing lots of it, as the costs and tradeoffs have greatly diminished. In fact, I’ve learned that building my own tools and frameworks, when done properly, comes with huge performance benefits and helps me understand the problems I’m trying to solve much more deeply. There has never been a better time for yak shaving!

I hadn’t thought about that, it’s a valid use case and likely to have increasing demand as drone deliveries become commonplace in the next few years.

Very convincing indeed! I really want to know what AI made that, as I’m looking forward to creating personal/customized music in a similar way.

I wonder why Google doesn’t bother competing with Microsoft in the flight simulation niche. All that Google Maps data would be pretty cool to use for that purpose, but instead we’ve got only this toy feature inside Google Earth.

I disagree. I’d prefer if my government invested more in AI solutions, so as not to depend so much on foreign technology.

In an ideal world, Brazil would have a thriving private sector, capable of competing even in the AI sector. Unfortunately, that’s not the case, and I believe that without government action such endeavors won’t really succeed.

I wouldn’t describe what happened here as incompetence. As a “carioca”, I am pleasantly surprised to know that the government’s IT department is involved in AI work — even without the budget to create its own models from scratch.

That “Don’t Insure Me” option hidden in the middle of a country list is pure evil. I’m used to seeing dark patterns everywhere but that’s a first for me.

From where I stand, it’s not fair to charge the hell out of people who fall for these tricks while giving steep discounts to the ones who don’t. Maybe there’s a “fool me once” aspect to Ryanair’s shenanigans, so at least their impact might be limited somehow.

Not OP but, anyway, AI output should be treated like any other source material.

I study from reputable sources every day and never cease to be amazed by how many errors or misconceptions they have. Peer-reviewed articles, books from renowned scholars, news from major publications… regardless of the source, false information and contradictions accumulate. I’d wager that AI, besides helping me uncover these issues in the literature, has had a lower error rate than most of the materials that I read on a daily basis.

Voice AI only feels natural if conversation moves at the speed of speech […] At OpenAI’s scale, that translates into three concrete requirements: Global reach for more than 900 million weekly active users

Surely the number refers to the total users of ChatGPT overall, and the fraction of those who use voice features is considerably smaller, is it not?

That’s the kind of thing that influences business decisions like knowing how much hardware and software optimization to throw at a problem.

Extremely high!

Those various caveats there — “value-aligned”, “safety-conscious”, “case-by-case agreements” — probably mean that no project ever will be “worthy” of OpenAI’s assistance.

In the unlikely event that an abiding project appears, then yeah, sure, it’s very probable that OpenAI would assist it :)

Five-figure minimum spend sounds pretty expensive for the vast majority of businesses out there. Of course, just a drop in the bucket for major brands.

Will we also see a GPT-5.5-Codex version of this model? Or will the same version of it be served both in the web app and in Codex?

I don’t understand why these bets are allowed at all. Can one just make an account there and bet in anything?

The whole “prediction market” charade is increasingly proving to lend itself to abuse. I hope regulations catch up with it soon, otherwise more shenanigans will follow.

I also liked the part about growing the company while reducing its carbon footprint by more than 60%.

Even if that figure might somehow be inflated, it is impressive nonetheless.

I don’t closely follow the news about Apple and now I’m wondering why they decided to go forward with this change at this moment.

As the world undergoes increasing supply chain issues, wouldn’t it be in Apple’s best interest to keep Tim Cook as CEO for a while? Or is he the one who’s looking to transition to a less demanding position?

Meta does both. It has long been said that businesses have little organic reach in Meta’s platforms, as an incentive for them to use ads.

I'd prefer to see board (or executive) level signatories over lay employees -- the people who can enforce enterprise policy rather than just voice their opinions

Although it would be nice to have some high-level signees there, I think we shouldn’t minimize the role of lay employees in this matter. Without having someone knowledgeable enough to build and operate them, AI models are worthless to the C-suite.

The Department of War is threatening to […] Invoke the Defense Production Act to force Anthropic to serve their model to the military and "tailor its model to the military's needs"

This issue is about more than the government blacklisting a company for government procurement purposes.

From what I understand, the government is floating the idea of compelling Anthropic — and, by extension, its employees — to do as the DoD pleases.

If the employees’ resistance is strong enough, there’s no way this will serve the government’s interests.

I agree. And even if those workers stay in the U.S., there’s absolutely no guarantee that they’ll do their best to favor the government’s interests — quite the opposite, if anything.

At the end of the day it’s a matter of incentives, and good knowledge work can’t simply be forced out of people that are unwilling to cooperate.

The problem with forcing public policy on companies is that companies are ultimately made from individuals, and surely you can’t force public policy down people’s throats.

I’m sure nothing good can come out of strong-arming some of the brightest scientists and engineers the U.S. has. Such a waste of talent trying to make them bend over to the government’s wishes… instead of actually fostering innovation in the very competitive AI industry.