ByteAtATime

I have essentially no knowledge in this field, but another concern may be that if the model has minimal cybersecurity knowledge, it may also implement insecure systems as a side effect.

> LLMs need an insane amount of text. Dessa learned language from hundreds of times fewer words (in fairness, LLMs also have to learn C++ and whatever the hell Angular is).

Now that a larger proportion of text is AI, I'm appreciating human-written text more and more; this got a good chuckle out of me.

Inconceivable!

Super interesting read; I especially found this letter from "P791" fascinating:

> Hi, Wonyoung, I map North Korea to help North Koreans.I select areas that look like they might be good points to escape. In particular, the detailed mapping around Phungso-ri (풍서리) is almost entirely my work.I mostly map along the Tumen River, both on the North Korean and Chinese sides. It is important to map the fences, roads, and towns on the Chinese side.

Interesting - I wonder if it's because coding doesn't use the specific token probabilities, while decision models do

I think, considering the size of this model, it's closer to a Pro than a Flash on everything other than speed

That makes sense -- I'll go figure something out, thanks for the explanation!

Fair point! I've actually thought about that before; I've tried to be extremely clear in the project's README with a disclaimer that this is a non-commercial hobby project and is not affiliated with the official Raycast team in any way.

The name is just for identification, as the project's goal is to be a compatible, open-source alternative for the Linux community, a platform they don't currently serve.

That being said, I'll definitely keep it in mind. Thanks for bringing it up!

One of the reasons people take notes is that you’re processing the information while taking the notes. This is removing that important step, and I would argue it means these notes are less effective than manually taking them in the first place

In my opinion, HDR is another marketing gimmick -- the average layman has no idea what it means, but it sounds fancy and is more expensive, so surely it's good.

Back in 2010, this "require[d] a decent graphics card"

Now, my phone's integrated graphics can run it very smoothly. Moore's law at play.

I think one of the main selling points of Cursor, as an investor-backed company, is that it's cheap. For $0.04 per prompt, I can get Claude 3.7 Sonnet to use 25 tool calls. In comparison, one of the images in the article shows either one prompt or a conversation that cost $7 (a third of Cursor's monthly subscription).

This is really cool! Just a small nitpick: on a low-powered device, the hero globe is really laggy (it's fine if I scroll past it, though).

Have you seen duckdb? https://duckdb.org/

It's basically what you're building, but more low-level. Really cool, to be honest -- serves the same market too. Do you have any significant differentiator, other than charts?