Vuizur

5 ms·

1.
by Vuizur 11mo ago
I think migrating away from office is not realistic for many companies. They run on Excel, and just breaking one vital table because it uses a macro can cost much more money than the entire office suite for many users over years. The Munich
2.
by Vuizur 1y ago
It is not very good at hard tasks, its ranking is much worse there.
3.
by Vuizur 1y ago
The next question is if LLMs are actually more sexist than the average human working in HR. I am not so sure...
4.
by Vuizur 2y ago
No, for this you need https://github.com/rmcrackan/Libation
5.
by Vuizur 2y ago
Grok-2 is rank 2 on LLM arena, it's basically as good as the best Gemini model. They already caught up. Only the latest ChatGPT model is a tiny bit better.
6.
by Vuizur 2y ago
I once wanted to compile a program that used Xapian on Windows. It was basically impossible for mortals. Imo people should use cross-platform alternatives.
7.
by Vuizur 2y ago
It will likely be amazing, Sam Altman said that the step between 4 and 5 will be like the one between 3.5 and 4. You can of course doubt him, but we'll see... I guess it will be this year, some guy working at OpenAI already posted &quo
8.
by Vuizur 2y ago
Wikimedia is unfortunately becoming one of the worst places to give your money to. They have their closed-source infrastructure, which now for years has not been able to generate HTML dumps without a significant percentage of articles missi
9.
by Vuizur 2y ago
There is also PyApp, which I think is really promising. The docs there are not that comprehensive yet and maybe a bit confusing, but the packaged programs usually work out of the box, unlike with pyinstaller.
10.
by Vuizur 2y ago
>executing large-scale changes in entire repositories in 3 years You can look at SWE-Agent, it solved 12 percent of the GitHub issues of their test dataset. It probably depends on your definition of large-scale. This will get much better
11.
by Vuizur 2y ago
It is impossible to remove personal data ("any information which are related to an identified or identifiable natural person") from the LLM training data. As far as I understand it ChatGPT and all other similar systems are blatant
12.
by Vuizur 2y ago
You can use DBeaver to get a good diagram for almost all RDBMs.
13.
by Vuizur 3y ago
Everybody who has released a decent model has used a shadow library dump for training. Are they going to admit that they illegally downloaded more than 20 million books?
14.
by Vuizur 3y ago
Scholz does not stall support for Ukraine, Germany has given about 10 times as much military aid as France to Ukraine: https://www.politico.eu/article/military-aid-ukraine-france-... Also compared to the US, Germany ha
15.
by Vuizur 3y ago
Reading the GitHub issue about this is somewhat entertaining: https://github.com/docker/for-linux/issues/690 People are getting hacked a lot because of this, and docker doesn't seem to care all that much
16.
by Vuizur 3y ago
Do you mean direct air capture (DAC) or carbon capture and storage from burning fossil fuels?
17.
by Vuizur 3y ago
He's talking about the new C# extension, Pyright and Pylance are free for everyone.
18.
by Vuizur 3y ago
The head of the Eclipse foundation said that they are aware, but that they thought that it would be too expensive for small businesses to get the verification. So the open source devs should do it for free. Related timestamp: https:/&
19.
by Vuizur 3y ago
When you look at open source AI, for example, EU laws are a huge problem: Not only the act you mentioned, but also GDPR, where it is not clear so far how the "right to be informed" or the "right to deletion" can/mus
20.
by Vuizur 3y ago
He apparently got divorced and moved out of his home country. And here he has no social circle. There are of course hopeless cases, but he is not mentally ill, only alcoholic. It is possible that he will never overcome alcoholism, but it di
21.
by Vuizur 3y ago
There are very few real-world-tasks where GPT-4 is worse at reasoning than the average person, right now only long-term memory, latency, TTS and STT makes virtual people suck. Once that is solved we should of course give them to old lonely
22.
by Vuizur 3y ago
Decent local LLMs such as Vicuna-7B are already able to run on phones with 8 GB RAM: https://mlc.ai/mlc-llm/ Right now it's a prototype, but in the future local LLMs will be everywhere.
23.
by Vuizur 3y ago
Currently many for-profit companies try to do the same, but additionally emotionally manipulate the user into giving them as much money as possible. (To not even mention automated catfishing...) This project is much better, but still might
24.
by Vuizur 3y ago
It is a community newspaper that is (apparently) independent of the WMF, and they don't have quarrels about hosting WMF-critical articles. Anti-WMF sentiment is surprisingly common on some Wikimedia projects.
25.
by Vuizur 3y ago
I think this applies to people with a university degree, it is rare for them to speak bad or no English. (Many speak it absurdly well.) But an average young nurse will probably only speak bad English and can maybe say a few broken sentences
26.
by Vuizur 3y ago
Similar to ChatGPT it also fails the "What is heavier, one pound of feathers or two pounds of lead?" test. So far only GPT-4 passes that one.
27.
by Vuizur 3y ago
I could replicate the refusal, but all it took to get the model to reply properly was replacing "evil hacker" with "hacker". Paternalism is not that big of a deal right now, especially when compared to ChatGPT/Bing
28.
by Vuizur 4y ago
I think it is a great idea. The problem is that institutions/decision makers have very little knowledge about open source software so it is really hard to convince them to do this. I can only speak about Germany, but I recently read a