punnerud

16 ms·

241.
by punnerud 2y ago
Cursor AI already have the option to switch to using claude-3-5-sonnet-20241022 in the chat box. I was about to try to add a custom API. I’m impressed by the speed of that team.
242.
Show HN: CSV SQL Graph – Graph visualization of views(github.com)
8 points by punnerud 2y ago|0 comments
243.
by punnerud 2y ago
Any statistics on the kind of people that stick to office work. A, B, or C players? Introverts vs extroverts, programmers vs project managers..
244.
by punnerud 2y ago
Is this the same page that had the background effect indicating level of the battery?
245.
Robert Cailliau – Collaborated with Tim Berners-Lee on the World Wide Web(en.wikipedia.org)
2 points by punnerud 2y ago|0 comments
246.
Show HN: Up to 2x more vertical video watching (Blank Black Screen, batterysave)(apps.apple.com)
1 points by punnerud 2y ago|0 comments
247.
Show HN: Camera Autodelete –Take pictures without camera roll, deletes in 30days(apps.apple.com)
1 points by punnerud 2y ago|0 comments
248.
by punnerud 2y ago
It’s starting to get way easier these days with the help of Claude.ai and Cursor.ai. I find it’s easier to pay much more focus to details when the iteration process is faster.
249.
by punnerud 2y ago
Different old songs (still on Shazam) in lat and long direction, if you find two of them you have the exact location. Could also place a directional speaker on top of your cars roof to not listen to it yourself or interrupt neighbors, just
250.
by punnerud 2y ago
The ones saying they will never use Python because it’s slow, is the probability high that their “fast” language is not thread safe?
251.
by punnerud 2y ago
As long as you have them you are allowed to keep a digital copy as well? Can you also rent out your digital copy?
252.
The Open Source AI‑Powered Code Editor(trypear.ai)
2 points by punnerud 2y ago|0 comments
253.
by punnerud 2y ago
Most likely not legal to invalidate in EU. There is laws that’s say that ev everything you can do manually, you are allowed to automate. Any rules against that is null and void. The “horse winning race” case is a known one where they go int
254.
by punnerud 2y ago
Isn’t the software for training end-to-end NN to be used in automation? Just a first version that it’s used for cars, and they have been using it for their own robot. So the claim still stands?
255.
Openpilot – Operating system for robotics(github.com)
241 points by punnerud 2y ago|140 comments
256.
by punnerud 2y ago
I have a new version that utilize the graph. Not pushed it yet. Then I use the embeddings to tag the answers and use tags + graph to try to understand if it is a good or bad reasoning. Hope to have it out next week. A bit to many bugs now.
257.
by punnerud 2y ago
I wish they could break Snapchat, Facebook etcs ‘s self-signed certs. I own the device, why can’t I see the traffic to and from all of these apps if I add my self-signed cert and approve to use a MITM-proxy. Most apps work, but not everyone
258.
by punnerud 2y ago
Switching to a low level integration will probably not improve the speed, the waiting is primarily on the llama generation of text. Should be easy to switch embeddings. Already playing with adding different tags to previous answers using em
259.
by punnerud 2y ago
Nice, thanks for the feedback. I have a prototype of also using the embeddings for categorizing the steps, with "tags/labels". Almost take it as a challenge to be able to reason better with a smaller modell than those >70B
260.
by punnerud 2y ago
Forgot that. Fixed now
261.
Knowledge graphs using Ollama and Embeddings to answer and visualizing queries(github.com)
122 points by punnerud 2y ago|9 comments
262.
by punnerud 2y ago
Maybe we could improve it more by combining it with embeddings? It’s a way to convert a text or response into an array of numbers, that can be used for similarity lookups. I made a way to query large datasets of text strings: https:/&
263.
by punnerud 2y ago
Worked, bud did not see a great improvement over llama:8b
264.
by punnerud 2y ago
I changed it into running 100% locally with ollama:8b: https://github.com/punnerud/g1 Not updated the Readme yet
265.
by punnerud 2y ago
Definitely AI, almost all of the gears does not work. A real artist already putting this much effort into it, would never make this mistake
266.
by punnerud 2y ago
Should be possible to both with the same database. Should maybe add that (did not have the need for it myself). Playing with the idea to add two intersecting streets and find the coordinate where they are crossing
267.
by punnerud 2y ago
Also recommend checking out this fork: https://github.com/BlackCatDevel0per/rgcosm Added CLI and import to Python
268.
RGCosm – Reverse Geocode for OpenStreetMap(github.com)
46 points by punnerud 2y ago|6 comments
269.
by punnerud 2y ago
I like the fact that Elon Musk read the draft… (credit at the end) Probably a great example of founder mode?
270.
by punnerud 2y ago
The problem is often that people use small MicroSD (8-16GB) that don’t support the number of writes to SQLite3 that HA does. Have been using Western Digital purple 64GB MicroSD for years without problems (support the same number of writes a

More ›