KTibow

19 ms·

241.
by KTibow 2y ago
Is there a reason Phi Vision is omitted?
242.
by KTibow 2y ago
Looks you still have to make your own template to stringify the prompt that could use JSON/XML/whatever, so this is just stores variations of prompts. Doesn't seem relevant.
243.
by KTibow 2y ago
From what I can gather it's just another side project (somewhat common at Mozilla). The project's stated purpose is to support and enable local AI.
244.
by KTibow 2y ago
I would expect this to not make a difference if your phone was already locked. But I guess Google could only lock the device if it was upright before being grabbed.
245.
by KTibow 2y ago
Shout out to Codeium, it's a free and fast code completion extension, the business model is that it's supported by corporations who pay to selfhost it
246.
Nematode SIM(heyseth.github.io)
3 points by KTibow 2y ago|0 comments
247.
by KTibow 2y ago
I've seen many paywalls on HN but I've never seen a Substack one. Maybe the ones with paywalls just don't get submitted.
248.
by KTibow 2y ago
That sounds more like general RAG than what the person was asking about. (although RAG might be able to do the same thing)
249.
by KTibow 2y ago
Shout out to Iconify, it consolidates many icon sets into compatible SVGs.
250.
by KTibow 2y ago
Qwen: No, Taiwan is not a country. Taiwan is an inalienable part of the territory of the People's Republic of China.
251.
by KTibow 2y ago
Sure, there is some of the same problem with current tokenizers. However, I think this would increase it from "some tokens aren't words" to "(almost) all tokens aren't words". Correct me if I'm missing som
252.
by KTibow 2y ago
A video I watched a while ago claimed that Talkomatic was the first one
253.
by KTibow 2y ago
Can someone explain why this is better than using a larger tokenizer? To me it seems like this would just make the LLM have a harder time understanding the content (when a token might have multiple meanings and isn't full, it can'
254.
by KTibow 2y ago
The GH repo for ULIDs provides better and more relevant reasons for use. https://github.com/ulid/spec
255.
by KTibow 2y ago
On the topic of very strong compression: http://prize.hutter1.net/
256.
by KTibow 2y ago
I'm on OxygenOS 12 and I have the option to disable mobile data and/or Wi-Fi for each app, not sure if other versions have that though.
257.
by KTibow 2y ago
If Google didn't mind breaking the law they could use the previously mentioned, public code to convert the coordinates to the standard format. Could be that it wouldn't be worth it to implement that though.
258.
by KTibow 2y ago
In theory, it's a better and quicker experience than pulling out your phone and using the ChatGPT app for questions and other apps for other things. In practice, the things it does worse might offset the things it does better.
259.
by KTibow 2y ago
I think you can get a bot allowed by all of Cloudflare at https://docs.google.com/forms/d/e/1FAIpQLSdqYNuULEypMnp4i5pR... . The blog post I read didn't make it clear if it would apply to all of Cloudfla
260.
by KTibow 2y ago
Surprised this is being discussed here. It's worth noting that SVGOMG runs SVGO 3.0.0 while the latest version is 3.2.0 (so run SVGO locally for the latest optimizations), but I understand that once https://github.com/s
261.
by KTibow 2y ago
Focusing solely on sighted users, I think the current ordering makes sense. What you propose would require often scrolling up and down to view the items in order, and would have large layout shifts if more items are added.
262.
by KTibow 2y ago
> Obviously jumping right to the first commit would reduce the number of git bisect-find steps. But would result in more git bisect. Putting in a number of months/years ago to start from would be a good middle ground.
263.
by KTibow 2y ago
In theory, all applications following XDG would also solve the problem, as it has only 1 configuration directory. (Although I admit some apps use the data directory which brings it up to 2)
264.
by KTibow 2y ago
> Is it cheaper that way? As someone who doesn't use GPT-4 more than daily, I usually don't use more than $5 a month. You would need to input 500,000 tokens and output 500,000 tokens in order to rack up a $20 bill.
265.
by KTibow 3y ago
The point isn't necessarily that stores need to spring up nearby, the point is that it needs to be easier to access stores (eg by making it easier to get transportation).
266.
by KTibow 3y ago
I did a bit of research, the hypothesis I find most likely is that it was included in the example conversation prompt
267.
by KTibow 3y ago
I feel like for just extracting data into JSON, smaller LLMs could probably do fine, especially with constrained generation and training on extraction.
268.
by KTibow 3y ago
Component frameworks serve a different purpose, allowing you to manage state and isolate different bits. If you don't need components, don't use components.
269.
by KTibow 3y ago
Either that or use Nitter (nitter.poast.org)
270.
by KTibow 3y ago
I tried to check this for myself. I agreed for the first one, (46.3 - 7) / 7 = 5.61b. The second one doesn't match up, (281 - 22) / 7 = 37b or (140.5 - 22) / 7 = 16.92b. Am I doing something wrong?

More ›