Posts
Tech Dev Notes
@techdevnotes · Sep 13, 2026
Apart from Grok 4.7, what are we even expecting from SpaceXAI in september
Elon Musk
@elonmusk · Sep 14, 2026
Grok 4.8, which is a 2.5T model trained with our new C++ software stack, will finish training this week and start RL
1.2K
1.1K
18.7K
3.8M
Maciej Mikuła
@maciekmikula · Sep 14, 2026
We’ve built the fastest training stack in the world.
Elon Musk@elonmusk· Sep 14, 2026@techdevnotes Grok 4.8, which is a 2.5T model trained with our new C++ software stack, will finish training this week and start RL
Elon Musk
@elonmusk · Sep 14, 2026
True
140
92
1.4K
85.8K
Ashok Elluswamy
@aelluswamy · Sep 14, 2026
As Jensen would say, “We are not a car”
Alex Rampell@arampell· Sep 13, 2026I just bought a new Model Y and can confidently say cars and Teslas are no longer the same product.
A car is like a pony or horse. I still like driving a stick shift, it’s fun!
Tesla with FSD is a transportation robot.
I had one of the first Model 3s in 2017. FSD was
Elon Musk
@elonmusk · Sep 14, 2026
True, a Tesla is a personal robot transporter
274
335
5.7K
236.2K
Denis Ulmer
@denispleiades · Sep 14, 2026
Meanwhile in Germany summon is limited to 6 meters, which is not even usable because you have to be so close to the car that it detects you as obstacle.
Elon Musk
@elonmusk · Sep 14, 2026
Hopefully, this can be addressed soon with regulators in Germany
135
100
1.3K
110K
Beff (e/acc)
@beffjezos · Sep 14, 2026
Elon should start his own German techno club in SF with crazy cyberpunk aesthetic as the ultimate revenge
Varunram Ganesh@varunram· Sep 14, 2026I really wish Elon expands on if he refused to enter Berghain or if Berghain rejected him and put a rest to all the "here's the club that rejected Elon" posts out there
Elon Musk
@elonmusk · Sep 14, 2026
I’ve already been to Berghain twice, but declined the third time when they had rainbow peace messaging in giant letters on their wall, which was lame
436
327
11.6K
460.2K
Dogan Ural
@doganuraldesign · Sep 14, 2026
If I had to trust one person build AGI, it would be Elon Musk.
Elon Musk
@elonmusk · Sep 14, 2026
Thanks ❤️
309
192
6.8K
155.9K
Adam Lowisz X Meetup 🇺🇸🇵🇱🇪🇺🇬🇧🇺🇦
@AdamLowisz · Sep 14, 2026
Elon, how close do you think Grok 4.8 will be towards achieving AGI?
Elon Musk@elonmusk· Sep 14, 2026@techdevnotes Grok 4.8, which is a 2.5T model trained with our new C++ software stack, will finish training this week and start RL
Elon Musk
@elonmusk · Sep 14, 2026
That will be Grok 5
558
358
7.4K
836.7K
Lueul
@itslueul · Sep 14, 2026
Agree with “PACE THE FRONTIER” in public. Keep training.
The 2.5T Grok 4.8 model is going to be better than the current non-public models from Anthropic & OpenAI.
I think Grok 4.7 will be in the same category as Fable 5.1 & Astra. That’s probably why he’s delayed the release twice.
Elon Musk@elonmusk· Sep 14, 2026@techdevnotes Grok 4.8, which is a 2.5T model trained with our new C++ software stack, will finish training this week and start RL
Elon Musk
@elonmusk · Sep 14, 2026
Grok 4.7 should be roughly on par with Opus 5.0, not 5.1. Better in some ways, worse in others. We need to fix multimodal performance.
Grok 4.8 will be a noticeable improvement.
Grok 4.9 is probably Astra/Fable class.
Grok 5 maybe better than anything. We shall see.
1.1K
975
15.3K
2.3M
Brivael Le Pogam
@brivael · Sep 14, 2026
Yes, and everyone should be able to create new stuff without any form of shitty regulation or patent.
Poverty will be fixed by engineering, not bureaucracy.
DogeDesigner@cb_doge· Sep 14, 2026Engineering is humanity’s greatest superpower.
Elon Musk
@elonmusk · Sep 14, 2026
True
151
104
1.6K
57.1K
DrKnowItAll
@DrKnowItAll16 · Sep 14, 2026
Wow. Is it looking good so far?
Elon Musk
@elonmusk · Sep 14, 2026
It’s unequivocally better than the 2.1T that was trained with Jax, but we made some mistakes that were only corrected mid run and we have greatly improved the data quality in the past few months.
The upcoming 3T run with improved internal training software and much better data
92
102
1.3K
148K
The Babylon Bee
@TheBabylonBee · Sep 13, 2026
'Well Done,' Says God To Man Who Spent Life Arguing In The Comments Section buff.ly/3OMGuDB
Elon Musk
@elonmusk · Sep 14, 2026
😂
178
90
2.2K
117K
Aaron Slodov
@aphysicist · Sep 14, 2026
your version of safety (reasonable) is not the same as the EAs/doomers, and they're using your name to drive their agenda
Elon Musk
@elonmusk · Sep 14, 2026
Yeah
140
80
1.7K
134.2K
Gad Saad
@GadSaad · Sep 14, 2026
This @jonstewart buffoon is truly insufferable. His gaslighting is Orwellian.
Winston Marshall@MrWinMarshall· Sep 13, 2026Unfathomable level of delusion from Jon Stewart
Elon Musk
@elonmusk · Sep 14, 2026
If there was a prize for “most skilled propagandist”, Jon Stewart would win
749
768
16.5K
232.7K
Brivael Le Pogam
@brivael · Sep 13, 2026
OMG.
If frontier AI is built by governments, we are completely fucked.
Don't do that.
Elon Musk
@elonmusk · Sep 14, 2026
True
342
234
6.2K
148.1K
Aaron Burnett
@aaronburnett · Sep 14, 2026
After the C+++ rewrite, our modeling indicated RL time for these models starting at 4.8 COULD be cut in half for ~2T models and substantially more time savings on a % basis for larger models. The thing I’d suggest watching is time between releases once they get some of the kinks
Elon Musk@elonmusk· Sep 14, 2026@itslueul Grok 4.7 should be roughly on par with Opus 5.0, not 5.1. Better in some ways, worse in others. We need to fix multimodal performance.
Grok 4.8 will be a noticeable improvement.
Grok 4.9 is probably Astra/Fable class.
Grok 5 maybe better than anything. We shall see.
Elon Musk
@elonmusk · Sep 14, 2026
Our training (aka pre-training) software is now the internally developed C/C++ stack, which was written primarily by the Starlink software team. This is awesome and probably best in industry.
However, our reinforcement learning stack is still somewhat of a legacy pile of many
265
328
6.6K
394.6K
Tom Elliott
@tomselliott · Sep 14, 2026
A week ago, no one was talking about AI-driven "extinction." Now, it's the only story.
Elon Musk
@elonmusk · Sep 14, 2026
I’ve been talk about it for an eon
939
289
7.4K
252.8K
JayViper
@JayViperTV · Sep 14, 2026
Can you please relax the over-moderation in @imagine? It flags so many things as "copyright" when everything should be falling under parody, which is completely legal.
Elon Musk
@elonmusk · Sep 14, 2026
Will investigate
134
59
912
81.6K
Sir Doge of the Coin ⚔️
@dogeofficialceo · Sep 14, 2026
I think about this a lot
Elon Musk
@elonmusk · Sep 14, 2026
Not wrong 😂
261
279
10.6K
176.6K
Patrick C Toulme
@PatrickToulme · Sep 14, 2026
The MFU on this full C++ stack is part of the reason xAI is churning out so many models. I would bet their stack is full of inline PTX and most likely completely AI generated by stronger models like GPT and Claude Fable.
The need for higher level frameworks for prod training is
NIK@ns123abc· Sep 14, 2026🚨 JUST IN: Elon says Grok 4.8 is a 2.5T parameter model that finishes training on SpaceXAI's new C++ stack this week and then starts RL
Elon Musk
@elonmusk · Sep 14, 2026
Our C/C++ training stack was written by humans. AI is not yet good enough for extremely high performance software (but it will be).
539
568
10.4K
1.2M
Paul D. Thacker
@thackerpd · Sep 14, 2026
Leslie Stahl alleges hiring pardon lobbyists in Trump adminstration is "a new system".
Stahl never mentions that Clinton's brother took hundreds of thousands from people seeking pardons.
Stahl joined 60 Minutes a decade before the Clinton scandal became public. @bariweiss
60 Minutes@60Minutes· Sep 14, 2026“There’s definitely a lot of money flowing around. And that makes a Petri dish for corruption.”
Law professor Mark Osler calls the new pardon system “anarchy.”
Elon Musk
@elonmusk · Sep 14, 2026
Hmm
123
95
1.2K
85K