Magnus Revang (@MagnusRevang) header
Magnus Revang (@MagnusRevang) profile picture

Magnus Revang

@MagnusRevang

Bluesky: @magnusrevang.bsky.social

 Earth
Joined December 2023
  • Tweets66
  • Following167
  • Followers30
1Media
Photo by Magnus Revang (@MagnusRevang)
So $KOSS just decides to rip on 5 x outstanding float in volume and +100% in the first hour of trading? On no news? The US stock market is the "best" market in the world, where no insider trading, manipulation or hidden short selling ever happens 馃憖
2
3
13
1,499
It's now +160% and almost 40m. It's not a short squeeze since SI is only 8% (unless it's possible to hide the SI, finra?), it's not a gamma squeeze since $KOSS does not have options. So... what does that leave? Somebody buying 9.001.000 shares of it? With a 5m free float? 馃槀
1
1
5
327
No more boomers in office. Period. 65 should be the maximum age for politicians. Senate congress president. All.
8
1
1
22
1,061
Forget Musk vs Zuckerberg in MMA, what I would pay to watch is Yann LeCun vs Musk in debate
Yann LeCun (@ylecun) profile pictureYann LeCun
@ylecun25 Jun 2024
I actually do have Elon's number. But we haven't spoken in a while.
117
Last generation open-source models was trained on 2-5T tokens, this generation is trained on 10-30T tokens. It will be really interesting to see the downstream effects of this 240T token dataset being available on the next generation of open-source models.
Rohan Paul (@rohanpaul_ai) profile pictureRohan Paul
@rohanpaul_ai24 Jun 2024
A 240T tokens dataset is now available for your LLM training. 馃く I don't even know how to go about downloading a 240T dataset lol. FineWeb's 15T comes out to 48 Terabytes. Can you imagine what a 240T looks like? 8脳 larger than previous SOTA (RedPajama-Data-v2 30T 125TB)
Image from Rohan Paul (@rohanpaul_ai)
58
Hear hear, I keep repeating this, Yann keeps repeating this, a lot of AI researchers keep repeating this - yet people keep listening to salespeople with an economic interest in you believing the opposite
Yann LeCun (@ylecun) profile pictureYann LeCun
@ylecun23 Jun 2024
Replying to @alex_peys
The problem isn't that it is a transformer. The problem is that it is an auto-regressive LLM. Auto-regressive LLMs that compute each token with a fixed number of computational steps can't reason, regardless of the details of the architecture.
1
78
If we are to avoid another AI winter, inference workloads need to grow 100x-1000x that of training workloads. And that makes the focus on nVidia as the only option strange.
Engineering Randomness (@EERandomness) profile pictureEngineering Randomness
@EERandomness21 Jun 2024
Replying to @robertgraham
You aren't seeing all the other tech NVIDIA has around the GPUs. Nvlink, infiniband, CUDA, etc. NVIDIA is objectively the best solution for training models by a wide margin. For inference, it is a closer race. Inference has dramatically different requirements.
51
Still do... calling it X (the company formerly known as Twitter) is just too long...
Not Elon Musk (@ElonMuskAOC) profile pictureNot Elon Musk
@ElonMuskAOC19 Jun 2024
When was the last time you called it "Twitter"? Be honest.
Image from Not Elon Musk (@ElonMuskAOC)
Image from Not Elon Musk (@ElonMuskAOC)
35
Wonder how many VC billions will be spent the next couple of weeks on this?
Ilya Sutskever (@ilyasut) profile pictureIlya Sutskever
@ilyasut19 Jun 2024
Replying to @ilyasut
We will pursue safe superintelligence in a straight shot, with one focus, one goal, and one product. We will do it through revolutionary breakthroughs produced by a small cracked team. Join us: ssi.inc
Safe Superintelligence Inc.

Safe Superintelligence Inc.

The world's first straight-shot SSI lab, with one goal and one product: a safe superintelligence.

ssi.inc
43
Load more