Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
sturza
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
10 ms
·
121.
▲
by
sturza
2y ago
The author has a very specific twitter feed. I would not generalize based on that.
122.
▲
by
sturza
2y ago
The feeling of stress comes from an inability to choose between obligations - greatly put
123.
▲
by
sturza
2y ago
It’s not the same on the phone. It’s async in the chat, whereas on the phone, the picking up is already accepting the conversation start.
124.
▲
by
sturza
2y ago
This.
125.
▲
by
sturza
2y ago
enshitification
126.
▲
by
sturza
2y ago
I got 2 Chromebooks 8-10y ago and never looked back. No maintenance calls, ever.
127.
▲
Can't Get No Satisfaction
(nymag.com)
1 points
by
sturza
2y ago
|
0 comments
128.
▲
by
sturza
3y ago
Hi, i’m doing something similar but rewriting books with 40-60% reduction. We should talk.
129.
▲
Gemini 1.5 and Google's Nature
(stratechery.com)
7 points
by
sturza
3y ago
|
0 comments
130.
▲
by
sturza
3y ago
Last time i build a real time news service i used the paid API of Inoreader, where i'd manage my own RSS feeds on their platform, and i'd poll their API every once in a while. Inoreader API would cost ~100 per year, not month. To
131.
▲
by
sturza
3y ago
In this repo: https://github.com/Jrbiltmore/UniversalBankAndTrust looks like he's talking to chatgpt to generate the code. It appears someone let an AI run on the internet.
132.
▲
Ask HN: Are there any enjoyable LLM based text games?
30 points
by
sturza
3y ago
|
5 comments
133.
▲
Mr. Boston Official Bartender's Guide
(en.wikipedia.org)
1 points
by
sturza
3y ago
|
0 comments
134.
▲
by
sturza
3y ago
This is great and all but notion is slow. After being a paying user for a number of years, i pulled the plug because of its atrocious performance.
135.
▲
Microsoft partners with Mistral in second AI deal beyond OpenAI
(theverge.com)
2 points
by
sturza
3y ago
|
0 comments
136.
▲
by
sturza
3y ago
I use my own RAG implementation to learn with. Contact me for a tool demo.
137.
▲
LoraLand: 25 fine-tuned Mistral-7B models that outperform GPT-4
(old.reddit.com)
2 points
by
sturza
3y ago
|
0 comments
138.
▲
LoRA Land: Fine-Tuned Open-Source LLMs That Outperform GPT-4
(predibase.com)
7 points
by
sturza
3y ago
|
0 comments
139.
▲
by
sturza
3y ago
And in non EU countries.
140.
▲
by
sturza
3y ago
It's asked to rewrite following certain guidelines.
141.
▲
by
sturza
3y ago
Updated link: https://filebin.net/29yn8bqtq6cx34ii
142.
▲
by
sturza
3y ago
https://filebin.net/29yn8bqtq6cx34ii
143.
▲
by
sturza
3y ago
Summaries are easier to generate, but making it shorter, still having the author's voice, is more appropriate for my goals. I've tried a lot of summary iterations and it turns out i don't want to read ABOUT the text. It'
144.
▲
Show HN: Book rewrite (reduction of 40-60%) service with LLMs
5 points
by
sturza
3y ago
|
9 comments
145.
▲
by
sturza
3y ago
Here's an example of 53% reduction of Walden, for your kindle: https://file.io/ynZg5xEzxAhD
146.
▲
by
sturza
3y ago
Let's give it a go, send me your book and i'll send you back the output. Or title and maybe i can find it online somewhere.
147.
▲
by
sturza
3y ago
I've built something that takes chunks of a book, in order, like 2-3 pages, and rewrites them in a more condensed format, like 40-60% of the original. Is this something useful to you?
148.
▲
Sweden's call for population to prepare for war sparks panic and criticism
(france24.com)
8 points
by
sturza
3y ago
|
2 comments
149.
▲
by
sturza
3y ago
I have similar interests, lets talk.
150.
▲
OpenAI Down
(status.openai.com)
1 points
by
sturza
3y ago
|
0 comments
More ›