arnaudsm

15 ms·

151.
by arnaudsm 2y ago
I calculated and found almost no correlation (r=0.2) between battery life and price. Which is quite sad considering it's the most important feature to consumers. Here's my write-up alongside some interesting dataviz : https:&#x2F
152.
by arnaudsm 2y ago
How do they compare to their original quants on ollama like q4_K_S?
153.
by arnaudsm 2y ago
My 16 yo cousins already use it and love it. They even showed their friends because the curation makes it a compelling entertainment source. And they request me specific stuff all the time, which makes me discover their culture too.
154.
by arnaudsm 2y ago
Students cheat because our society values diplomas more than knowledge. If we reversed that belief system, cheating would disappear
155.
by arnaudsm 2y ago
American's trust in media declined accordingly. https://news.gallup.com/poll/651977/americans-trust-media-re... My point was about small communities though. For country scale organizations it's a much tr
156.
by arnaudsm 2y ago
People want cigarettes and fast food too. It's not good for them either, and legislation and education improved the situation. Social media is to the 2020s what cigarettes were to the 50s
157.
by arnaudsm 2y ago
In a small communities, newcomers are more scrutinized to prevent that. And the scammer can only have a single victim before being expelled. HN, while being large, incentivizes that brilliantly, by highlighting new accounts in green.
158.
by arnaudsm 2y ago
Pseudonyms are anonymous yet can still build trust. I appreciate anonymity when I want to talk about sensitive topics.
159.
by arnaudsm 2y ago
Reputation is the usual solution. You build a community and you know who to trust, like humans have always done. The early internet was mostly that (BBS, Usenet, forums). But on the modern internet we mostly consume random google sites &amp
160.
by arnaudsm 2y ago
Homework caused major educational inequalities anyway, I'm not sad to see it disappear.
161.
by arnaudsm 2y ago
I've seen hundred of college students successfully cheat with mobile phones in class.
162.
by arnaudsm 2y ago
There's a difference between accidentally stumbling on shocking TikTok content and willingly browsing a gore site. If my kid really wants to check it, it might be acceptable, but I'm not ok with an algorithm showing it by surprise
163.
by arnaudsm 2y ago
Yes. Before the internet, my parents looked at the PG rating of every movie we rented at Blockbuster. Many parents did.
164.
by arnaudsm 2y ago
This is great for creating a media-garden for your children! Kid content on YouTube and TikTok can be extremly disturbing, even for an adult. Until they turn 16 I intend to vet most media myself on a self-hosted server.
165.
by arnaudsm 2y ago
I'm tired of LLM releases that cherry pick benchmarks. How does it compare to SOTA qwen2.5/phi3.5 ? Anyone knows an up to date independent leaderboard? Lmsys and livebench used to be great but skipped most major models recently.
166.
by arnaudsm 2y ago
Thank you, will contact you asap, I'd be happy to help :)
167.
by arnaudsm 2y ago
OpenAI's code quality leaves to be desired, which is surprising considering how well compensated their engineers are. Their recent realtime demo had so many race conditions, function calling didn't even work, and the patch suggest
168.
by arnaudsm 2y ago
Brave's shady crypto strategies [1] and history of injecting headers and affiliate codes in your navigation [2] makes it an untrustworthy browser to me. [1] https://coinfomania.com/brave-browser-users-monthly-bat-rewa..
169.
by arnaudsm 2y ago
Same here, is this the legacy voice mode with text-to-speech only ?
170.
by arnaudsm 2y ago
I tried, it solved a few steps and was quite impressive! It's really good at riddles and associating random concepts. It got stuck when it had to calculate directions though.
171.
by arnaudsm 2y ago
What's the endgame of a dead internet? Everyone leaves and most interactions happens in private group chats? It's the serendipity of the original internet I'll miss the most.
172.
by arnaudsm 2y ago
No, because it was proven to be a forgery. Russian metadata was left by accident in the dump. Check the files yourself if you don't believe it. https://www.theguardian.com/world/2017/may/08/macron-ha
173.
by arnaudsm 2y ago
That's incorrect, the metadata of the emails were incorrectly scrubbed and proven to be forgeries. You can check the files yourself. https://www.theguardian.com/world/2017/may/08/macron-hackers...
174.
by arnaudsm 2y ago
LLM intelligence has plateaued for the past 12 months. Open source models are catching up while being 20x smaller than the original gpt4 (gemma2/llama3.2/qwen2.5). AI companies are promising AGI to investors to survive a few more
175.
by arnaudsm 2y ago
It's not in the public interest if it's not authentified. We had a similar leak during the french presidential election, the "MacronLeaks". It was a mix of fabricated scandalous data, and legitimate boring data. I'm
176.
by arnaudsm 2y ago
I was comparing expert systems here, not emergent ones
177.
by arnaudsm 2y ago
"earth copy" is just marketing lingo. Models are approximations made for a specific goal at a time. You cannot build a model that is both good for gaming and meteorology. 99% of that engine value is procedural terrain and graphica
178.
by arnaudsm 2y ago
Is there an up-to-date leaderboard with multiple LLM benchmarks? Livebench and Lmsys are weeks behind and sometimes refuse to add some major models. And press releases like this cherry pick their benchmarks and ignore better models like qwe
179.
by arnaudsm 2y ago
You should also try phi-3-small 7B, seems much better at reasoning according to https://livebench.ai
180.
by arnaudsm 2y ago
The latency of Groq is impressive, much better than o1! Did you benchmark your system against MMLU-pro?

More ›