stolsvik

7 ms·

31.
by stolsvik 3y ago
Yeah, but if they wanted, they could have AI’ed this to death. https://x.com/stolsvik/status/1746351452618981887
32.
by stolsvik 3y ago
This is obviously not an isolated incident. Bitcoin University has put a comment on their video about the problem, and Michael Saylor of MicroStrategy (who is being impersonated) has published several xeets about the multiple scams, both hi
33.
YouTube putting scammy ad about "double your Bitcoin" on Bitcoin University vid(twitter.com)
14 points by stolsvik 3y ago|9 comments
34.
by stolsvik 3y ago
Thanks! Both for the work, and the answer! The “.. on to other, more important things” left me wondering! Would it be possible to shed some light on that?
35.
by stolsvik 3y ago
Huh, why? Internet access is close to a human right these days, IMHO.
36.
by stolsvik 3y ago
“The Diamond Age: Or, A Young Lady's Illustrated Primer” https://en.m.wikipedia.org/wiki/The_Diamond_Age
37.
by stolsvik 3y ago
Code reviews. Some “interesting” code, which actually work, but which is inefficient in multiple ways. Dig deeper and deeper. Architecture suggestions for a relatively simple problem, discuss the suggestions.
38.
by stolsvik 3y ago
Worth noting that all reviewers have voted reject: https://openreview.net/forum?id=MGWsPGogLH But it would be nice to see an implementation of it.
39.
Turing complete transformers: Two transformers are more powerful than one(openreview.net)
6 points by stolsvik 3y ago|3 comments
40.
OpenChat-3.5-1210: Best Open Source 7B LLM(twitter.com)
3 points by stolsvik 3y ago|0 comments
41.
by stolsvik 3y ago
I have no formal credentials to say this, but intuitively I feel this is obviously wrong. You couldn’t have taken 50 rats brains and “mixed” them and expected the result to produce new science. For some uninteresting regurgitation, sure. Bu
42.
by stolsvik 3y ago
You should really try it. I’ve been coding professionally since 2000, and some 10+ years before that as a kid with a hobby. The AI takes away quite a bit of tedious stuff. When you think “I’ll just have to code out that annoying little dumb
43.
by stolsvik 3y ago
I agree: https://news.ycombinator.com/item?id=38313501
44.
by stolsvik 3y ago
Whether Greg knew of the decision beforehand? I don’t think so. This totally business-as-usual post from Greg Brockman happened 1 hour before the one from OpenAI: https://x.com/gdb/status/1725595967045398920 http
45.
by stolsvik 3y ago
This totally business-as-usual post from Greg Brockman happened 1 hour before the one from OpenAI: https://x.com/gdb/status/1725595967045398920 https://x.com/openai/status/172561190026258
46.
by stolsvik 3y ago
So, since we’re all spinning theories, here’s mine: Skunkworks project in the basement, GPT-5 was a cover for the training of an actual Autonomous AGI, given full access to its own state and code, with full internet access. Worked like a ch
47.
by stolsvik 3y ago
The “restart from where it failed”-aspect was a big reason for why I made Mats3. It is message-based, async, transactional, staged stateless services, or message-oriented asynchronous RPC. Due to the transactionality, and the “state lives o
48.
by stolsvik 3y ago
Worth checking out his "ChatGPT AutoExpert" prompts too, one of which is "Developer Edition" utilizing the python environment of Advanced Data Analysis. https://github.com/spdustin/ChatGPT-AutoExpert
49.
by stolsvik 3y ago
On many Linux DEs you have “Alt-drag” and “Alt-resize”. It is absolutely amazing, and luckily there are hacks to get that on Windows too.
50.
by stolsvik 3y ago
How many megabytes of text is it reasonable to expect a user to read? This could have been interesting, rendered meaningless by overflow of legalese.
51.
by stolsvik 3y ago
Sources?
52.
Llama-2-long: 32k context length(twitter.com)
3 points by stolsvik 3y ago|0 comments
53.
by stolsvik 3y ago
As far as I know, the device I’m writing this on is also made in China. And I don’t even have to specify the brand.
54.
by stolsvik 3y ago
Interesting. I found a random news article from 28th January 2022 about Australia committing 1B AUD to the reef: https://edition.cnn.com/2022/01/27/australia/australia-great... I asked 3.5 and 4 “Austral
55.
by stolsvik 3y ago
Those are the same: https://adoptium.net/ “Eclipse Temurin is the name of the OpenJDK distribution from Adoptium.”
56.
by stolsvik 3y ago
That was really good. At first I thought it was a bit dull with just those 5 images, but the tons of different styles and concepts exemplified made it a great inspiration.
57.
by stolsvik 3y ago
> _green thread inefficiencies_ Ron Pressler (@pron) from Loom @ Java had an interesting talk on the Java Language Summit just recently, talking about Loom’s solution to the stack copying: https://youtu.be/6nRS6UiN7X0
58.
by stolsvik 3y ago
Counterpoint: I don’t refer to 3.5 when I say ChatGPT. I pay for ChatGPT, and always use GPT-4. Which I believe every paying customer do.
59.
by stolsvik 3y ago
Me too, but then I realise that one should include at least a few tech savvy friends - or references to good coworkers or similar which these friends could contact.
60.
by stolsvik 3y ago
Don’t you also measure the current of the “inner probes” (the ones “driving a current”), to verify that you do have contact?!

More ›