Sem Day (@SemDayy) header
Sem Day (@SemDayy) profile picture

AI models, rockets & frontier tech šŸ§ šŸš€ | daily takes on what just dropped | memes when the news deserves it

Seattle, WA
Joined September 2026
Photo by Sem Day (@SemDayy)Photo by Sem Day (@SemDayy)Photo by Sem Day (@SemDayy)Photo by Sem Day (@SemDayy)Photo by Sem Day (@SemDayy)Photo by Sem Day (@SemDayy)Photo by Sem Day (@SemDayy)Photo by Sem Day (@SemDayy)Photo by Sem Day (@SemDayy)
hey, I'm sem šŸ‘‹ I follow AI and space every day, from new models to starship launches. dream is to study in switzerland and work at an AI lab. building a community of people obsessed with the frontier. drop a comment with what you're working on
3
67
self reported productivity is the weak spot here. the number that matters is whether those 1,800 researchers actually published more or got results faster. if the follow up data shows that, this becomes the strongest case for AI in science so far
Scientists reported nearly doubling their productivity with AI in our study of more than 1,800 participants in the ChatGPT for Academic Researchers program. Using privacy-preserving methods, we looked at how they use ChatGPT across their work—from developing ideas to analyzing data and writing code—and what it could mean for scientific discovery. openai.com/index/science-at-…
1
28
catching the booster was the headline, but catching the ship is what makes starship fully reusable. once the upper stage comes back intact, cost per launch stops being about hardware and starts being about turnaround time. also ship 42 with don't panic on it is perfect
Don't Panic Riding on Starship 42 during final approach to catch
1
2
47
the no training on program data part matters more than the number. most labs can't paste unpublished research into a model without that guarantee. going from 10k to 200k researchers in under 4 months says the first wave actually used it
We’re expanding free access to our frontier models for up to 200,000 scientists worldwide, with training and support. Applications open Nov. 23. By default, we don’t use data shared through the program to train our models. openai.com/index/expanding-c…
2
20
everyone in these replies is asking for the same thing in different words. more access, faster releases, less waiting. the cyber team asked what to improve and the answer is simple, let people actually use the good stuff
Cyber team wants your feedback on Daybreak. Models, codex plugins, blue/red access, documentation, red teaming & defense factory... What can we do better or improve?
1
13
month 1 on X. 780k impressions, 66 followers. turns out going viralin replies and building an audience are two completely different games. here's what i'm changing
1
12
tried gemini 4 carbon for a full day (7-8 hours). it's insanely good at planning butistill went back to claude for actual code. not sure google wants to hear that
1
17
invite only, no deliveries yet, and you can't even use it as a cab. that's not publicly available, that's a beta test with wheels. mkbhd's hair is winning this bet so far šŸ”„ but never count elon out in december
The memes are everywhere, so let’s go back over the bet: Cybercab publicly available for under $30k by the end of 2026 My hair is safe for several reasons 1: No deliveries yet. But they’re estimating starting in December so that’s neat 2: Invite only, not actually publicly available yet 3: You can’t actually use it as a cab(??!) It’s limited to private use only. Inside the existing geofenced locations. It’s a paperweight if I take this thing home. That’s one’s not actually part of the bet but yeah not exactly what was promised in this first wave Hey @grok, set a timer for 48 days from now
1
84
30k upfront, 300 a month, one car per person, only in texas, invite only and estimated delivery december. tesla made the most exclusive subscription in the world and it has wheels šŸ”„ can't wait for the full review
Alright chat I’m in. Here’s the deal Cybercab order page is live. Invite only. Must accept delivery to one of these cities in Texas. Only operates in those existing geofenced areas. Requires a $300 monthly subscription to operate. One vehicle per customer. ā€œEstimated deliveryā€ December
1
47
Google didn't just beat @AnthropicAI or @OpenAI . It beat both at once. Gemini 4 Hydrogen, not even their most powerful model, tops Claude Sonnet 5.5 and GPT-6.1 Terra on 7 of 10 benchmarks. Huge respect @Google šŸ”„
1
244
Gemini 4 Hydrogen leads 7 of 10 benchmarks vs Sonnet 5.5 and GPT-6.1 Terra, and it's the fast, cheap tier. Claude still wins terminal work and computer use, but this gap is getting very small
1
77
Google didn't just beat Anthropic or OpenAI. It beat both at once. Gemini 4 Hydrogen, not even their most powerful model, tops Claude Sonnet 5.5 and GPT-6.1 Terra on 7 of 10 benchmarks. Huge respect @Google
1
84
24/7 in every city is huge @robotaxi but which city is next, and when are you coming to europe?
Robotaxi now runs 24/7 Every city we're available in
3
170
AI launches don't feel like launches anymore. They feel like a guessing game One-word posts, cryptic teasers, Polymarket odds, and model names that sound like the periodic table. Argon, Carbon, Astra, Fable, Ember Honestly, I think the hype is moving faster than the products. Half the models people argue about aren't even in the apps yet. You read the benchmarks, open the app, and it's still the old model. Meanwhile every lab is teasing the next thing before the last one has actually reached anyone I don't care who posts the best teaser. Whoever wins next year will be the one that ships something people can actually open and use on the day it's announced
2
90
The long-term vision is fascinating. No landing legs, no unnecessary mass, just a system designed around reusable infrastructure on Earth and eventually on Mars
1
55
less repeating myself is the upgrade i actually want. @finkd whats a task you trust muse with now that you wouldnt have trusted it with last week
Upgraded the models for Muse and Meta Al, bringing Active Memory and Active Focus from Muse Wonder to everyone. In Muse, these new specialized models are 2.4x less likely to forget critical info and are 67% better at self correcting and completing everyday tasks which means less repeating yourself and fewer dead ends. More on the models: research.meta.ai/blog/specia…
2
50
my macbook still works perfectly and somehow apple made that feel like bad news 😭 @johnternus what would you show someone who thinks they dont need to upgrade
New Macs and iPad mini available today!
3
50
realtime muse voices is actually wild @finkd can you switch voices mid convo or do you pick one and stick with it šŸ‘€
Now in the Muse app: talk with your Muse in real time. And pick from multiple voices šŸ—£ļø
2
75
I don't feel like I need nano texture on the ProLED MacBook Pro fwiw I thought it was a must-have for the MiniLED panel though
1
153