acangiano

Because they shipped and are better only for specific applications. For everything else, they lag behind frontier models. They win in latency-sensitive niches (e.g., autocomplete, code infilling, subagent calls). Not frontier-level reasoning.

I use both but I agree that they are generally not on par. I find Claude Code does a better job and doesn't overengineer as much. Where sometime Codex does better is in debugging a tough bug that stumps Claude Code. Codex is also more likely to get lazy and claiming to have finished a large task, when it reality it just wrote some placeholder lines. Claude has never done that. They might be on par soon, however, and I think Anthropic is playing a dangerous game with their limit enforcement on people who are on subscriptions.

If you can swing it, I'd go for the Bulova Lunar Pilot or the Seiko Prospex Chronograph SSC819. If you are looking for cheap Chinese, Phylida makes a well-reviewed one.

There are a few brands that have been punching above their price points. For Rolex Submariner homages, you can't really beat Sugess. Other Chinese brands worth considering are San Martin, Cronos, WM, and Seagull. Among the upcoming brands, Watchdives and Tactical Frog are also doing quite well. Though cheaper, Pagani Design is a step below the rest. It's not bad, but their watches are definitely not as polished as the ones from the brands mentioned above. In terms of movements, stick to NH35, NH34, PT5000, SW200, Miyota 90S5, etc. Try to buy from the official stores on Aliexpress, and check your private messages after you add the watches to your cart. They'll often issue coupon codes for said items.

Salaries are consistently above other mainstream stacks, partly because most Elixir jobs look for senior engineers.

For graphing you can use: https://hex.pm/packages/vega_lite

This is only one source among the many you should consult, also I'm biased because I'm a co-founder, but cognitiveclass.ai constantly publishes new guided projects (and courses) on related topics. They are free, and in the case of guided projects, quick. Sort by new and have at it: https://cognitiveclass.ai/courses?type%5B%5D=all&sort%5B%5D=...

I second this. The phone format really makes me pick it up more.

I'm getting less and less impressed with the company, so I started diversifying. I got Linux installed on my desktop and started using it more often. Switched from Apple Music to Spotify. Got rid of my Apple Watch Ultra and switched to Garmin. Went from Things to TickTick since it's cross platform. From Notes to Notesnook for the same reason. I know it's a drop in the bucket and I'm still pretty invested, but they are steps towards independence should I decide to abandon their ecosystem altogether at some point.

In a recent scandal, salmon fillets wrapped around lead weights shoven down the fish's troats was also used to cheat.

The chicken breasts were fairly large 386g cooked (so even more raw). 315g of green beans. When I provided the grams it did better but was still off by 250+ calories (I’m guessing it ignored me mentioning that that was cooked weight.) It’s a cool tool and it can be useful but if you’re not careful you’ll run into some very inaccurate guesstimates. One way to improve accuracy would be for the system to learn from previous logs (both in the system and previous apps.)

Pretty cool but unfortunately it’s wildly inaccurate. I said “two chicken breasts and lots of green beans”. It estimated 262 kcal. The “actual” calories were 754 kcal. Specifying the actual grams was better, of course, but still off by quite a bit.

The UX for the analysis part is actually significantly better on Chess.com

Troy Hunt called that behavior "way too far into the grey for my comfort" in a recent post about the massive Naz.API leak.

Silver lining to taking it after my heart attack.

Stories like this would be a game changer for this new company. I genuinely hope they go for it. In a way, it's surprising that there isn't a charity dedicated to this type of effort for drugs considered safe.

Natural aging will impact you much faster than AI.

A large percentage of the population going vegan and working from home becoming the norm for suitable jobs, would be game changers in terms of pollution.

It was the right call because even if you were overly cautious, you can always go back again.

Any app in particular?

It appears to be the fate of such package manager creators. The homebrew guy was also rejected (by Google).

25 minutes you'll be okay with never getting back.

I have four new interns who recently started in my team. I'm setting them up for blogging and made the same point to them (almost verbatim).

Another thing that died is the practice of linking to each other like it's going out of style. These days the majority of backlinks are earned through aggressive SEO techniques, not organically like 10-15 years ago. This penalizes regular blogs in favor of crappy blogs run by SEO experts.

I still believe in blogging and I think that the lesser competition, in a roundabout way, makes it easier to stand out. However, discoverability in Google can be a frustrating process both for readers and publishers.

Also, 2% is very dubious. Italy is at 7%. The Philippines at 8%. I mean, sure these figures are out of people who were tested, but 2% is looking awfully optimistic at this point. So the potential for death reduction is even greater than the intentionally unassailable 1% figure you picked.

This is a very interesting model to monetize open source and it's not limited to that. It could be used for ebooks, courses, videos, art... anything, really.

The fact that they open it up to everyone once the desired sponsorship goal is reached is a nice touch provided it doesn't make sponsors who paid feel shortchanged.

One way to prevent that is by acknowledging sponsors upon release. The other way, somewhat naturally occurring in most cases, is the delay required to achieve the target number of sponsors.

Generally, I like it.

My favorite trick is to not ask trick questions.

Amusingly, Twitter itself could use one. My account (@acangiano) has all of its images censored under "sensitive content" even though they are 100% benign images. No amount of tweets to @TwitterSupport has done anything at all to change it. There is basically no recourse. My account is like 12 years old and has 4.5K followers, so it's not like it's a random spam account, either. ¯\_(ツ)_/¯

My mental picture of her is that of a genius on the floor drawing on huge pieces of paper.

It comes from this short article about her: https://www.nytimes.com/interactive/2017/12/28/magazine/the-...