obert

5 ms·

1.
by obert 3y ago
don't want to sound dismissive, it's known that llms understand state, so you can couple code generation + state, and you have sort of a runtime. E.g. see the simulations with linux vm terminals: https://www.engraved.bl
2.
by obert 4y ago
I noticed some odd results when switching from "today" view to "this week" view. Maybe they take into account some hidden parameters like diversity of the audience, time distribution, reputation, or maybe they have an od
3.
More than 80% trending repos are about AI(github.com)
3 points by obert 4y ago|2 comments
4.
SKonsole: Streamlining PR and Commit Message(github.com)
2 points by obert 4y ago|0 comments
5.
by obert 4y ago
sure they did, but they can’t show
6.
by obert 4y ago
FWIW dupes could be detected, would be a nice HN feature in fact, grouping similar news. Not free, ok, but doable.
7.
by obert 4y ago
it’s been 20+ years some doctors keep bringing up a correlation between gut bacteria and MS, and I don’t see anything new here, again. Always hoping for the best, but the trend is not promising well, at this pace it will take another 100 ye
8.
by obert 4y ago
and when time comes and AI can run on a laptop, Apple will be 3-4 years ahead of everyone else…
9.
by obert 4y ago
same here, we’ll be watching for variations of avg humanity IQ, with very big differences between the impact on kids, adults and elders
10.
by obert 4y ago
if they dont fix it on MacOS and Linux they’ll keep getting bad feedback. The whole point of using web tech is being cross platform, looks like they don’t get it
11.
I wanted to try the new Bing with ChatGPT. Then, Microsoft went all Microsoft(zdnet.com)
5 points by obert 4y ago|0 comments
12.
by obert 4y ago
The writing was on the wall: when looking for info we don’t interact with people continuously distracted by ads. We ask, focus on the answer, act. Google business model is shortsighted when looking at tech in terms of decades of humans mode
13.
by obert 4y ago
https://rewind.ai right?
14.
by obert 4y ago
You can double those numbers if you consider all additional costs.
15.
by obert 4y ago
Uh, I didn’t notice it because on iOS it’s a static image at the top of the page
16.
by obert 4y ago
though PR should explain Why more than What, ie I can figure out what changed looking at files, but without context and design details an autogenerated description doesnt make a review any easier
17.
by obert 4y ago
the problem with all techniques used to protect from prompt injection is that they are not explicitly asking the model. Always validate input before processing it.
18.
by obert 4y ago
Screensaver https://www.screensaversplanet.com/screensavers/what-colour-...
19.
by obert 4y ago
“Windows bridge to Android” would have been a better choice
20.
by obert 4y ago
I moved away too, this happened while I was closing all my wallets across multiple plats
21.
by obert 4y ago
personal experience, few bucks fortunately, lost and I’m giving up on it
22.
by obert 4y ago
- python: https://huggingface.co/docs/transformers/model_doc/gpt2#tran... - javascript: https://www.npmjs.com/package/gpt-3-encoder - c# https://github.com/dluc/open
23.
by obert 4y ago
Binance is a scam. They make it extremely hard to withdraw money, with errors, fake limits, unresponsive customer support.
24.
by obert 4y ago
? what
25.
by obert 4y ago
ChatGPT is in fact just a chat prompt on top of Davinci3, plus a markdown renderer
26.
by obert 4y ago
the chat supports some sort of markdown, so you can do other similar things, e.g. create documents, tables, etc.
27.
by obert 4y ago
what you’re probably looking for is becoming an AI app developer, leveraging inference to create new types of products. No need to become an expert in ML and training, to start with. That’s similar to what some people did about 12-15 yrs ag
28.
by obert 4y ago
yes, T=0 means no randomness, and given sufficient tokens the output should always be the same (in such case whether they cache is a matter of hit ratio, I'd guess)
29.
by obert 4y ago
In general it's better asking questions after the content, and T=0.7 is way too high. Here's HN homepage summarized with T=0 and TopP=1, asking "Summarize the content above.": DV2> The content above is a list of the t
30.
by obert 4y ago
have you tried tweaking the behavior, e.g. asking not to guess, to be brief, providing examples, constraining output with syntactic sugar, etc? e.g. it's like talking to a person, give feedback and you might get better result.

More ›