AI + tech. Dayjob: Nuclear power Prev: Grok / xAI

Virginia, United States
I resigned from xAI tonight. It makes me very sad, but was the right thing to do -- and here's why. xAI told me I either had to delete the post quoted below, or face being fired. After reviewing everything and thinking a lot, I've decided that I'm not going to delete the post -- which is very clearly a harmless personal opinion. Why did they tell me to remove this opinion? Well, according to them, the reason is that I acknowledged that Grok 3... exists. I wish I was joking. I'm not. That's the reason -- the fact that I wrote "Grok 3 (TBD)" is grounds for being fired. But wait, hasn't Grok 3 been officially acknowledged by xAI? Yes. Yes it has. I'll post below the official xAI blog post talking about Grok 3, along with many public Elon posts and video where it is repeatedly acknowledged. To be clear, the post they wanted me to remove is 100% just my personal opinion. I do not know where Grok 3 will stack up against other SOTA models. Hopefully it does well, I don't know. That's why it says "opinion" and "to be determined." It will probably be pretty good at some things and imperfect at others. I didn't think this was a particularly wild opinion. Again, their official demand said that even writing "Grok 3 - TBD" is somehow "confidential information." This is absolutely absurd, since it's repeatedly been acknowledged by the company and its famous CEO. Are they mad that my clearly-labeled opinion didn't guess that the still-unreleased Grok 3 will be higher? Maybe. Probably. Again, maybe it is at the top, I genuinely don't know. That's why it says "to be determined." The specific feature of Grok I spent the majority of my time working on with a really hard-working team is very cool and I hope it works extremely well for everyone. I won't say what it is because that would be **actual** confidential information. (Maybe after it comes out.) I still hope Elon and xAI win. Yet...... It's very disappointing to me that a company and leaders who supposedly champion free speech and openness would try to fire a low-level employee over a clearly-labeled opinion that contains absolutely nothing controversial, but here we are. The entire situation has been very strange. I thought about just deleting the damn thing.... But you know, once you start caving and giving up holding mild personal opinions, the slope becomes very slippery. I'll keep my speech and dignity and get another job, or build one. Catch ya on the flip side.
The ranking currently (my opinion), for code: ChatGPT o1-pro o1 o3-mini (all kind of tied) Grok 3 (expected, tbd) Claude 3.5 Sonnet DeepSeek GPT-4o Grok 2 Gemini 2.0 Pro Series (might be higher, will probably move up)
2,983
1,329
21,612
6,302,257
X app is the best app, example #57372
91
1,266
20,333
1,441,670
If we were a proper country.........
79
390
16,661
334,507
I just searched "how many 4090s do I need to run DeepSeek locally" and ya'll shorting NVDA
228
283
11,038
1,053,068
It turns out "the next Google" was also Google
71
563
11,215
268,830
Starlink essentially leap-frogged and replaced the entire terrestrial internet and cell networks This is still completely underappreciated and not priced in
323
451
9,736
743,273
"omg it's so crazy Starliner speakers are picking up phantom garbled signals!" people over 30:
140
375
8,872
251,759
When you have all of YouTube as training data:
"A pair of hands skillfully slicing a ripe tomato on a wooden cutting board" #veo
81
283
8,892
556,898
GPT-4o "free" on iPhone........ how does OpenAI make money from this?
1,144
225
8,446
2,963,552
Guys I'm sorry but this is dumb.
364
160
8,113
540,798
Confirmed: ChatGPT is now "stealth searching" the Internet in the background, without showing this to the user.
221
274
7,217
2,249,661
What if we ALL started using X via VPNs but set the location to Brazil Overwhelm them with digital red herrings to chase
495
654
7,269
630,668
Meta Orion AR glasses. Explain to me why Apple didn't do this? Steve Jobs would have.
729
281
6,991
916,077
188
555
7,019
52,264,830
How do you defend against this? (not a rhetorical question.....)
AI Notkilleveryoneism Memes ⏸️
4,629
745
6,877
1,930,826
Can someone explain why UBI wouldn't just become the de facto inflation level? Like, if you gave everyone $1,000 a month guaranteed, why **wouldn't** all prices just rise? If you get a "free $1,000" but your rent also goes up $1,000 (or food, etc) what is accomplished?
1,662
171
6,605
1,229,545
I am often critical of OpenAI but the Sora video previews have left me speechless. This is excellent, world-changing work.
400
652
6,475
6,414,139
Replying to @BrandonButch
This looks like every Android from ten years ago.
49
83
6,306
598,262
Replying to @anammostarac
Someone with an actual IQ of 80 would probably not be able to write intelligently, which the post clearly is.....
122
11
6,178
368,702
Replying to @bryan_johnson
"Hey just checking on your nighttime erection quality"
17
31
5,603
160,560
OpenAI has brought the director of the NSA to its board of directors Because of course
Paul M. Nakasone brings world-class cybersecurity expertise to OpenAI’s Board of Directors, helping us deliver on our mission by protecting our systems from increasingly sophisticated bad actors. openai.com/index/openai-appo…
231
1,039
5,333
878,019
AI Johnny Cash sings "Barbie Girl"
174
1,134
5,341
1,388,424
My xAI team has not taken a day off in 12 days They could have, they just didn't
298
135
5,059
443,184
Google current profitability: 34% net margin Google current cash on hand: $98.5 billion Yeah.
My conspiracy theory: Sundar Pichai wants the bubble to pop because it will hurt OpenAI a lot more than it will hurt Google.
54
148
5,128
484,300
Elon pays OpenAI $200 a month for Pro?
There you have it
412
77
4,680
422,920
Wait, the Android team is... one guy?
Grok is seeing tremendous growth in India for our Android app, developed by our 1-man team of @attilablenesi. We are looking to add the second engineer to the team. Apply here: grnh.se/209e0b367us
161
114
4,583
451,481
Elon has deleted his comment saying he didn't know Yacine was fired.
110
60
4,454
1,183,144
A private U.S. company just landed on the Moon. We are so back
101
275
4,281
623,539
Question: why would key people leave an organization right before it was just about to develop AGI?
415
126
4,260
1,018,425
Replying to @AdrianDittmann
Yes, well, can a 4090 do THIS
33
28
4,354
163,249
Trump had a chance to start the second term on a more serious, prepared tone but totally trashed that opportunity by launching the scammiest crypto shitcoin in history days before inauguration. Long sigh......
557
350
4,106
198,918
ChatGPT (Premium, GPT-4) now says it can't help you with code scripts. WTF is this?! Hey OpenAI... can you please explain why customers should continue paying for totally nerfed services? Because you're just sending all your users to your competitors at this point.
498
268
4,131
2,014,966
even the gpu rich are gpu poor
37
116
4,276
145,946
Replying to @jennyzhangzt
Can you explain how you're working on a PhD in AI yet just discovered undersea internet cables exist? This was nowhere in your previous education? I'm confused.
67
41
3,867
152,804
Google's head of "AI Responsibility" is apparently named Jen Gennai (Gen Gen AI)...? Simulation is getting weird.
228
261
4,051
855,209
I am the most junior person @grok My DMs are open. I will read every message. I will forward the most important to @elonmusk and the team. We will make Grok great. It matters. It is important.
559
176
3,998
479,045
China has a huge advantage in the AI race: total disregard for IP laws They will just train huge models on every movie, every song, all of YouTube We will complain and they will laugh and keep doing it I don't have an answer to this but it's a reality we must face now
413
297
3,611
392,348
Mexico just elected its first female President. She's 61 and has a PhD in Energy Engineering. Mexico about to accelerate?
411
149
3,468
298,403
Replying to @sama
The buried lede here is that you TOLD it to "pick one." Grok did as you requested. ChatGPT did not, ignoring the user request.
51
38
3,453
245,970
Keep a dumbbell next to your desk and do curls while waiting for Codex and Cursor to finish
133
127
3,434
153,723
Replying to @sgodofsk
This map is a good reason FOR the senate. Why should a rancher in Wyoming want to be told what to do by somebody in Manhattan? The senate is recognition that different places have different views yet are also valuable -- not least of which is because they make all the food.
81
58
3,122
72,021
I have accepted a roll at @xai to help train Grok. I am a very small part of this important endeavor. I hope to learn as much as I can. I hope truth wins.
313
81
3,244
165,706
Meet the new computer user interface..... a black, talking circle.
204
145
3,075
190,934
If OpenAI's o1 can pass OpenAI's research engineer hiring interview for coding -- 90% to 100% rate... ......then why would they continue to hire actual human engineers for this position? Every company is about to ask this question.
307
244
3,114
464,196
Prediction: Ilya will ship nothing, regret not joining Zuck or Elon, and be largely forgotten to history. I'm sorry I don't make the rules
209
46
3,170
378,383
Why are the aliens / "enemy drones" running standard port / starboard nav lights
807
123
2,931
400,720
Uhhhhh GROK?! 😆😁
142
24
2,906
2,202,811
The ranking currently (my opinion), for code: ChatGPT o1-pro o1 o3-mini (all kind of tied) Grok 3 (expected, tbd) Claude 3.5 Sonnet DeepSeek GPT-4o Grok 2 Gemini 2.0 Pro Series (might be higher, will probably move up)
365
200
3,022
5,498,324
Replying to @uncledoomer
Would you say you've...... hatched a plan
3
3
2,855
110,659
Here's the official statement released by Scarlett Johansson, detailing OpenAI's alleged illegal usage of her voice... ...read by the Sky AI voice, because irony.
225
492
2,774
723,021
ChatGPT, Claude, and Perplexity -- three totally different AI services -- are all down right now. What the actual F is going on?
738
235
2,706
751,220
This is huge. Recent academic paper shows that GPT-4 lost significant capability in code skills over a few months. I love GPT-4 but it would be nice to have OpenAI address this. They've been insisting that users are "imagining it," basically.
200
435
2,679
2,197,385
POV: Cursor "fixing your code"
Stefan Spain
76
190
2,897
318,691
How is this going?
111
150
2,706
166,459
How does it being "the thinnest product" help me create, Tim? Apple is lost.
Meet the new iPad Pro: the thinnest product we’ve ever created, the most advanced display we’ve ever produced, with the incredible power of the M4 chip. Just imagine all the things it’ll be used to create.
363
55
2,592
443,684
We are so fucking back
21
85
2,677
116,784
Sam this is called an API
an idea for paid plans: your $20 plus subscription converts to credits you can use across features like deep research, o1, gpt-4.5, sora, etc. no fixed limits per feature and you choose what you want; if you run out of credits you can buy more. what do you think? good/bad?
48
47
2,695
153,156
Microsoft CEO on OpenAI: Doesn't matter "if OpenAl disappeared tomorrow." "We have all the IP rights and all the capability." "We have the people, we have the compute, we have the data, we have everything." "We are below them, above them, around them."
170
249
2,599
686,173
Replying to @greg16676935420
greg's mom:
30
11
2,598
66,942
Steve Bannon just said Elon "isn't an American?" I was pretty neutral on this but that's fucking retarded.
416
124
2,540
156,271
> claims they basically have AGI > can't get their reminder app to work
Seems like a lot of you are enjoying this beta and we're working hard to scale up, but we are experiencing server issues right now and some tasks will not complete. We hope to have it resolved soon.
89
80
2,571
182,131
If you ask Gemini to remove an image background (make transparent png), it gives you the output in a JPEG which doesn't support transparency 🤦
94
34
2,644
190,531
Replying to @DudespostingWs
This isn't a "supercar" It's an n ADD Nova kit car... based on top of a 1970s VW-Beetle. Most Novas were built on a VW Beetle floorpan and used the Beetle’s air-cooled flat-four—typically the 1.6-liter (Type 1) twin-port making roughly 50 bhp, with some cars also running the 1200/1300/1500 Beetle engines. A few were later swapped to other engines, but the stock VW 1.6 was by far the usual choice.
87
36
2,509
282,977
So the "AI generated girl" in the car video? Yeah, she appears to be a 100% real person named Arial who does gig videos on Fiverr Some of the biggest AI influencers on X have been telling everyone this was AI generated, and two different AI video companies including @HeyGen_Official have claimed their tech generated them. This whole thing is really strange
299
283
2,315
1,340,724
Prediction..... Tim Cook's tenure at Apple will not be remembered particularly positively. Mostly taking safe paths and pushing odd products which don't sell well. -Do people want an Advanced Pencil? .....doooo they? -Siri AI... where is it? This should have dominated a year ago. Missed opportunity. -Vision Pro, being cut back because of lackluster sales. Almost all buzz has died very fast. -Apple Car project, killed after billions spent, and right as real Autonomous driving is reality -iPhone and iPad, nearly identical to what they were a decade ago Tim seems nice. But lacks... Fire? Maybe I'm wrong. Just a hot take.
Pencil us in for May 7! ✏️ #AppleEvent
540
91
2,293
997,909
I had to decline a job offer recently from ((major defense contractor)) because they wanted me to sign a paper that ANYTHING I CODED OR DEVELOPED even on weekends unrelated to the job became their IP if they decided.
I have been suspecting for a long time that employers are abusing the interview process to obtain free work. bc wtf do they mean "accepted into our codebase"
139
100
2,448
119,535
The slow realization that Yann is a petty, nasty, jealous little b!tch.
204
55
2,289
116,576
Alright I'm gonna say it... This is essentially AGI. This will be seen as magic to masses. What else do you call it when a virtual "person" can listen, talk, see, and reason almost indistinguishably from an average human? Isn't that AGI?
Say hello to GPT-4o, our new flagship model which can reason across audio, vision, and text in real time: openai.com/index/hello-gpt-4… Text and image input rolling out today in API and ChatGPT with voice and video in the coming weeks.
441
145
2,221
797,783
"OpenAI is nothing without its people"
117
162
2,217
147,163
Okay I take back my earlier criticisms Sora 2 is incredibly addicting and fun, this is a new era.
80
187
2,306
267,819
Shoutout to the brave Optimus who was able to get this Indian Ocean shot 🫡
68
73
2,265
65,126
"What is my purpose?" "You pass the butter."
102
133
2,106
263,297
What changed? (genuine)
Replying to @cb_doge
SpaceX will lean in big on the Moon
564
53
2,230
394,143
Wait so.... Sam's original "note" earlier today only addressed Mira leaving. Then he deleted it and posted the new one (up now), where he groups them all together and says "it made sense to now do this all at once" But... then why did he post the first note with no mention of the others. He didn't know? He had no idea three of his key people including the Chief Technology Officer were going to quit within hours of each other...?
59
71
2,148
360,751
I asked Sora for 3 Navy SEALs infiltrating a port. Close enough
127
93
2,173
95,807
If OpenAI is right on the verge of AGI, why do prominent people keep leaving?
279
67
1,618
283,178
Anthropic missing $ millions by not having a button that says, "buy more messages"
but when the world needed him most, he vanished
63
31
2,045
85,024
Would you use X as your main email, with a @x.com email address? 👀
Interesting. We need to rethink how messaging, including email, works overall.
371
42
1,990
188,333
GPT-4V vision jailbreak confirmed using PENGUIN. Fascinating! 🐧 (interesting details in comments)
48
208
1,970
422,824
Starlink essentially leap-frogging all internet and cell providers is a huge coup. Almost everybody was skeptical of Starlink a few years ago and it was panned as a dead-end vanity project with low chance of profitability. Now it's poised to be Earth's everything network.
121
77
1,879
130,945
No. No no no no. Stop. This contributes nothing to humanity.
24
4
1,702
148,493
Replying to @GrantSlatton
This is the Family Guy Italian meme but Japan 😄
9
1,868
72,497
If Doom never happened, no Quake. If Quake never happened, no GPUs. If no GPUs, the AI revolution doesn't happen. (This is of course a slight exaggeration but it's interesting to think about.....)
349
108
1,859
194,435
It is noteworthy that Microsoft is essentially nowhere in the current AI conversation.
277
48
1,857
183,524
OpenAI just shot Rabbit in the face.
60
84
1,771
143,315
What did Ilya Karpathy Logan Jan allllll see?
140
73
1,766
433,891
If you walked through a Best Buy a decade ago and today, there isn't actually much difference in tech The iPhones still look the same, the drones are about the same, the flat screen TVs are a bit bigger But there is no OH WOW! technology for sale that didn't exist 10yrs ago
257
54
1,865
100,713
Please stop sharing "AI enhanced" or "AI upscaled" images of the blurry FBI shooter photo. It literally made up a face. This does not help find the actual suspect. To fill in missing data, AI algorithms are guessing. Example:
66
109
1,793
120,470
I have no idea who the CEO of Anthropic is because he / she isn't constantly giving media interviews bragging about how amazing they are They just go to work and ship like it's their job. Weird.
50
85
1,684
105,999
Don't even scrub it, just run it. This goes hard.
42
54
1,649
108,888
This is so scammy.
I am extremely proud of what we continue to accomplish in crypto. $Trump is currently the hottest digital meme on earth and I truly believe that @WorldLibertyFi will revolutionize DeFi/Cefi and will be the future of finance. We are just getting started! 🚀🚀
104
85
1,641
64,067
"Grok doesn't do code hurrr durrrrrrr" Yeah, okay buddy. Skill issue or you literally didn't try. Here's Grok-written pong game, player vs computer. Took a couple minutes to make. 100% Grok 2 code.
201
180
1,596
6,580,773
Replying to @sog_on_bird_app
So to do more capitalism we have to be more socialist Yeah pardon if I'm really skeptical of this plan
82
5
1,634
65,453
GPT-4 has become so @#$%@ lazy, it won't even output more than a handful of code lines now: Fix. This.
196
77
1,625
482,862
By the the way, the answer is: DeepSeek-R1-Distill-Qwen-32B model requires around 74 GB of VRAM, which can be achieved using four NVIDIA RTX 4090 GPUs. At current prices, RTX 4090 cards range from $1,800 to $3,100 each, bringing the total cost for four GPUs to approximately $7,200–$12,400, depending on the retailer and model.
172
62
1,684
233,083
This is an actual code output ChatGPT-4 just gave me. FIX. THIS. ALREADY.
343
54
1,526
481,655