First impressions on DeepSeek v3 LLM and the future role of China in AI

[@fjkldhakljf]

Thank you for recording this in english 🥰

[@ncpeaksean4278]

That's helpful, thanks for explaining more about this LLM 🚀

[@VivianPatanio]

Ciao Salvatorem buon Anno! Very impressive video! Wishing the best!❤

[@antirez]

@@VivianPatanio thanks! And buon anno :)

[@zackmanrb]

Excellent English, thanks for making this!

[@DelShott]

A little bit off topic, but you mention that you took a paper and did an implemention. Is this something you do often? Are there any remarkable paper that led to a fun implementation journey?

[@antirez]

Absolutely! I do this often. I did this also for skiplists backthen. In general data structure papers, when well written, are very cool to implement. Another one that was quite cool to test was the initial eventual consistency paper from Amazon. Also: HyperLogLogs.

[@willv88]

7:08 The Chinese are soon going to lead with EVs despite close ties with Tesla in some cases. This is also highly impacting the German (and EU) economy. Any way, I agree with you. Very interesting share.

[@lucarossi1508]

Ciao grazie , buon anno! L’ho provata mi sembra impressionante..

[@ozonehephzibah6949]

Hello Salvatore, are you accessing the model Locally or you n the Cloud??

[@antirez]

@@ozonehephzibah6949 DeepSeek official web interface. Very well made.

[@mitchellchristianson8120]

First the amount of power we save at some point it'll become free. Would it not divided to the the cost of the bill already in power consumption?

[@fiblan00]

Beautiful video and interesting insights. I haven’t delved into DeepSeek v3 yet, but nowadays, discussing Open, LLMs, and AI in general is becoming increasingly complex. For a few months now, we’ve had OSAID (by OSI). What does it mean to have an open LLM today.... and tomorrow? I’d love to see a video of yours sharing your thoughts on this topic. ;)

[@privacyhelp]

Luigi?

[@mitchellchristianson8120]

I remember somebody saying don't build a rocket if you're in space already open AI same concept and here is a perfect example. Is it not? Deepseek

[@jamesrav]

I was impressed using it, although on one question it repeated the same line over and over and probably would not have stopped. I had to click for a new question to stop it.

[@alessandro-desimone]

Nice one Salvatore! I'm curious. How did you start learning about ML and LLM? Ciao :)

[@antirez]

Hi Alessandro, I followed the field of NNs starting from 2003 and never stopped, I find it to be one of the most interesting things in CS and human technology.

[@PeterNduati-f1q]

Dude just read your bio in your personal website...led me here...new subscriber...thanks for redis...I wanna do a clone of it because am in uni and am trying to learn C ..wish me luck​@@antirez

[@leosmi1]

DeepSeek is sick.

[@asbestomolesto]

Whoa, I'm the first one! :)

[@vincenzogiambusso]

Questo è il primo al quale do in pasto il manuale di un gioco. Gli chiedo una cosa molto specifica, che è sparsa in mezzo alle pagine, e mi dà la risposta giusta. Direi che è già tanta roba rispetto agli altri che ho provato.

[@antirez]

The "needle in the haystack" problem is solved very well by DeepSeek v3, like in Claude Sonnet. GPT4 context window has something very odd instead.

[@mitchellchristianson8120]

What does it mean? Ask them how they sleep at night knowing what this man did bring him to your next board meeting. I'm surprised they didn't jump out the window. It's a pretty big story and a lot of money

[@mitchellchristianson8120]

To understand 😢 in my life remember buster Douglas and Mike Tyson the 49ers against Denver they won the super bowl and they didn't even have to compete. They laid down and knew they'd been beat. We know that song. Charlie Daniels one of the all-time kick your ass quickly and now they move to the driving. Thank you, xpeng chip your own ship and sanction that

[@giuseppeambrosio555]

Nice🇮🇹🇮🇹