hk__2

It turns out that my issue was using high/xhigh. Going back to medium or even low made it 'intelligent' again.

Yes sorry I meant NASDAQ 100.

Ok but there are very few indices following NASDAQ, compared to S&P 500.

Yes. In my experience, I need to at least read a book for 1-2 hours to really enter the story, and then I can read for 10 minutes here and there. The issue is if I start reading the book 10 minutes by 10 minutes I never really enter the story and it takes me more time to finish it.

Neither the author, you or me have to wait for someone to ask us; we just share our thoughts with strangers in the hope to bring something positive into the conversation.

That’s addresses in the blog post:

One way to remove backtracking is to lift the pen immediately instead of waiting until the end of the word, as if doing italic calligraphy. Pen lifts alleviate the mental queue problem and give a chance to readjust the palm, but they break the writing flow.

First of all, you don’t have to make time to read. What you need to do is read every single time you are not doing something else.

Mmh I’m not sure about that. I prefer to read for 1-2 hours rather than read 2 minutes here and 5 minutes there, especially for books that require some concentration to read, like dense stories and/or books not in my native language.

"Snarky commenter" is not either.

Come on, "Elon" is the exception.

Her name is von der Leyen. "Ursula" is fine for your friends, but it’s awkward for a public figure. Do you write "Donald" as well?

The comment I was responding to was talking about Codex usage in the past few months. This is a general feeling about Codex with Claude, not a model-to-model comparison.

I’ve experienced this with GPT but not with Opus/Fable.

There _is_ a difference in the way Claude and GPT write. Last Friday I felt Opus was becoming dumb because it was writing like GPT.

I’d argue the opposite. I’ve switched back and forth from one to the other and Opus/Fable has been constantly better than any GPT in my daily work. It’s a bit slower but it does the things right, with as little code as possible, some comments where needed. Codex is faster but you always have to correct it because it got something wrong; it writes tons of code ("let me add a small helper") with obvious comments.

The test suite is the result of these years of years of running in production. Every time you fix a bug, you add a non-regression test to ensure you don’t break it again.

Sign up for Axios AI+ to continue reading for free.

Some explanations from https://clojure.atlassian.net/browse/CLJ-2961:

Clojure’s idiomatic use of maps has proven valuable, but missing required keys, misspelled keys, and invalid values can lead to failures that do not connect to the actual source of the problem (e.g. NPEs) making diagnosis difficult. At the same time, Clojure lacks a simple inline mechanism for functions to document and check the keys they require and accept. Existing tools either separate those expectations from the function itself or couple data shape and data provision.

Of course they do.

As long as emojis are used as an _addition_ to text labels I think this is a good thing. The problem arises when they are used _instead_ of labels, although that’s a problem that predates emojis, it’s common with normal icons.

If you design an interface when some actions are only behind icons/emojis (no text, no hover title), expect users like me to click on them just to see what they do.

Sure… but which ones? How can you know ahead of time?

Experience, mostly, and bit of trial-and-error. For code I use SOTA, but for e.g. translations I use Gemini 3.5 flash, for some other use-cases I use Gemma 4.

No you don't; it's often overkill to use the SOTA models. People want SOTA because it's shiny, but there are a lot of tasks where it's cheaper and more efficient to use other models.

it can be tricky to enforce but I think doable

It’s not doable at all. There are millions of people that don’t need a domain but would be happy to be paid $5-10 by some random scammer to hand over their domain.

Our rule of one person per subdomain will hopefully prevent this at scale

No it won’t. Spammers will just pay thousands of random people in poor countries to create their domain.

its tagged as satire at the very top of the page, first thing under the title

Not the first thing, it’s buried in the tags as grey on light grey on white.

The outrage is not that LLMs hallucinate; it’s that a search engine presents you with a response to your query that contains wrong information, and hides in a small, hidden-by-default menu that it’s LLM-generated.

Different article, but accessible to non-subscribers: https://www.bbc.com/news/articles/c3ryj81ywlro

Fixed link: https://gitlab.com/here_forawhile/torum

In the French translation of War and Peace I read, the first use of every diminutive had a footnote explaining who was that person.

I read War and Peace recently, is it the same amount of characters? At some point I almost started a genealogical tree of the characters.

OpenAI, the non profit organization

No, the nonprofit org stays nonprofit, while the for-profit org it owns will become publically traded.

See https://openai.com/index/evolving-our-structure/