imjonse

15 ms·

211.
by imjonse 2y ago
Mistral even used the term 'truly open' in their announcement: https://mistral.ai/news/mixtral-8x22b/ Also from the Llama 3 announcement: 'Today, we’re introducing Meta Llama 3, the next generation
212.
by imjonse 2y ago
GPT-5 will make elections obsolete :)
213.
by imjonse 2y ago
Opus was taken over by quite a few Gemini and GPT4 models on the chat arena leaderboard, hopefully this entry will put Anthropic back near the top. Nice work!
214.
by imjonse 2y ago
Seconded. He has very good videos on many topics and a down to earth attitude.
215.
by imjonse 2y ago
They have released the models https://huggingface.co/togethercomputer/Llama-3-8B-Dragonfly...
216.
by imjonse 2y ago
Both Leike and Sutskever are still credited in the post.
217.
by imjonse 2y ago
"From one perspective, Mamba-2 isn’t strictly better than Mamba-1: while it’s a dramatic improvement from a training perspective, Mamba-1 might be better from a pure inference perspective. Since inference speed of SSMs is entirely gove
218.
by imjonse 2y ago
Use edge to download Firefox ;)
219.
by imjonse 2y ago
So one of the main good aspects of Windows is that it can run Linux :) ?
220.
by imjonse 2y ago
It is not the first or the last company where people may despise their CEOs character but still stay for the advantages.
221.
by imjonse 2y ago
He rubs elbows with very powerful people including CEOs, heads of state and sheiks. They probably want 'one of them' in charge of the company that has the best chances of getting close to AGI. So it's not his technical chops
222.
by imjonse 2y ago
This is what llm cmd does for any command, and it also offers a chance to edit it before running https://github.com/simonw/llm-cmd
223.
Hard Prompts Category in Chatbot Arena(lmsys.org)
1 points by imjonse 2y ago|0 comments
224.
by imjonse 2y ago
Main main gripe about Helix is that helix-like keybindings in VSCode are not working well for the times I need to use that :)
225.
by imjonse 2y ago
Is there new evidence for this? It has been the main hypothesis for why the pyramids are far from the river, I thought it was generally accepted.
226.
by imjonse 2y ago
Including more basic (as of 2024) features in the core like Helix would indeed be a good idea but would go against the existing culture of vim with many plugins competing.
227.
by imjonse 2y ago
Oh I didn't mean that. I think it's the best benchmark, just it's not necessarily representative of ordering in any domain apart from generic human preference. So while Llama3 is high up there, we should not conclude for exam
228.
by imjonse 2y ago
Fortnite's revenue is more than 5 billion annually. This fine corresponds to about 2 hours loss of revenue.
229.
by imjonse 2y ago
Human preference does not always favor the model that is best at reasoning/code/accuracy whatever. In particular there's a recent article suggesting that Llama 3's friendly and direct chattiness contributes to it having
230.
by imjonse 2y ago
Telegram had some weird primitives which they said we should trust because they were made by their top team of mathematicians. Signal builds on widely used crypto primitives even if their protocol is their own (vetted by actual cryptographe
231.
by imjonse 2y ago
Durov's exile and distancing from Russia after the VK takeover may be just for show and for selling Telegram as 'the dissident app'. It is popular, easy to use and insecure.
232.
by imjonse 2y ago
Telegram were claiming they were more secure even when they had their own home-rolled crypto. Security is not Telegram's strong point and it never was.
233.
by imjonse 2y ago
The key is cheaper device combined with deep learning.
234.
by imjonse 2y ago
With all due respect for his academic accomplishments, confidence in this domain in the current climate is usually a signal towards potential investors; it can be backed by anything between solid work (as I hope this turns out to be) and a
235.
by imjonse 2y ago
Math notation has different fonts, with similar goals as syntax highlighting. It also works well in black and white :)
236.
by imjonse 2y ago
Should all arxiv papers be backed by economic considerations or business plans?
237.
by imjonse 2y ago
The "why for" is usually learning/gaining experience/FOMO.
238.
by imjonse 2y ago
Anything more open than OpenAI models can now call their models 'truly open'. It's good they will have recipes but they also don't seem to want to share the actual data.
239.
by imjonse 2y ago
They should not, but then they also should not call the model truly open. It is the equivalent of freeware not open source.
240.
by imjonse 2y ago
Phi-3 is instruct tuned though so hopefully better.

More ›