Gael Ollivier

502 posts

Gael Ollivier

Gael Ollivier

@gaelollivierdev

https://t.co/lurrnfCzjX

Los Angeles, CA Katılım Ekim 2012
78 Takip Edilen105 Takipçiler
Gael Ollivier
Gael Ollivier@gaelollivierdev·
@Erdayastronaut @thecyberfam Tim, you’re the goat, the OG, the one and only. The space community is so lucky to have you. Hater’s gonna hate. Your whole brand is “being a real human with real feelings watching a rocket launch and getting excited about space”. This is what we’re here for!!
English
0
0
5
418
Everyday Astronaut
Everyday Astronaut@Erdayastronaut·
I’m going to be honest. This entire thread is the thing I like least about my job. CLEARLY my motivation for a scrub was selfish in that I wanted to be there and felt like I was missing out on something I (and you all) have waited 7 months for. I also was being realistic about the odds of it going off on the first attempt. I was trying to temp expectations to help ease everyone (including myself) in the event of a scrub. That’s the reasonable thing to do as opposed to hype people up for a “certain” launch. I was keeping track of their timelines and given hold capabilities to try and gauge along with everyone if it would launch, again, a reasonable thing to do. I talked over Nikki Manaj because it had nothing to do with the mission as I do when there’s anything that’s a random tangent in any stream as we always do. Gave her a chance to speak, was surprised and confused to see her, thought it was funny, then didn’t think we needed to hear her asking when it was launching 🤷‍♂️ We’ve ALWAYS shared our footage with anyone who asks and especially our fellow creators. Ask anyone, @NASASpaceflight @CSI_Starbase @MarcusHouse @FelixSchlang Obviously I don’t just do this for money or else I wouldn’t spend 7 months on a video or every dime on a production trailer. This whole threads feels like a huge attack on my character and I’m not going to stand for it. I work my ass off, I’m proud of the work I do, I do right by our team and those who work with us. Just because I was sad I couldn’t be there doesn’t mean anything more than that.
English
333
92
4K
99.1K
Gael Ollivier
Gael Ollivier@gaelollivierdev·
@joebarnard Vibe code your own thing. It’s so easy to build an app like this with Claude and you can tailor it to your exact needs and workflow
English
0
0
0
55
Joe Barnard 🚀
Joe Barnard 🚀@joebarnard·
I'm trying to upgrade my project management. Between making rockets, all their parts, and then videos about that process, I need a tool that helps me keep track of these to-do lists and timelines. What do y'all use?
English
94
5
295
28.1K
Gael Ollivier
Gael Ollivier@gaelollivierdev·
@karpathy This is the kind of experiment that proves that even if models stopped improving, we’d keep seeing an increase in compute demand because there’s still so much to unlock with more inference compute!
English
0
0
0
338
Andrej Karpathy
Andrej Karpathy@karpathy·
ah yes, this is what post-agi feels like :) i didn't touch anything. brb sauna
Andrej Karpathy tweet media
English
73
59
1K
189.9K
Andrej Karpathy
Andrej Karpathy@karpathy·
nanochat now trains GPT-2 capability model in just 2 hours on a single 8XH100 node (down from ~3 hours 1 month ago). Getting a lot closer to ~interactive! A bunch of tuning and features (fp8) went in but the biggest difference was a switch of the dataset from FineWeb-edu to NVIDIA ClimbMix (nice work NVIDIA!). I had tried Olmo, FineWeb, DCLM which all led to regressions, ClimbMix worked really well out of the box (to the point that I am slightly suspicious about about goodharting, though reading the paper it seems ~ok). In other news, after trying a few approaches for how to set things up, I now have AI Agents iterating on nanochat automatically, so I'll just leave this running for a while, go relax a bit and enjoy the feeling of post-agi :). Visualized here as an example: 110 changes made over the last ~12 hours, bringing the validation loss so far from 0.862415 down to 0.858039 for a d12 model, at no cost to wall clock time. The agent works on a feature branch, tries out ideas, merges them when they work and iterates. Amusingly, over the last ~2 weeks I almost feel like I've iterated more on the "meta-setup" where I optimize and tune the agent flows even more than the nanochat repo directly.
Andrej Karpathy tweet media
English
330
553
6.5K
647.1K
Gael Ollivier
Gael Ollivier@gaelollivierdev·
@Erdayastronaut Have one and it’s my favorite hoody by far, love the quality and the nerdy details. Can’t wait for new designs!
English
0
0
0
30
Everyday Astronaut
Everyday Astronaut@Erdayastronaut·
I've heard a few people ask for us to put our space shuttle hoodie back in the shop... just trying to see if we have enough demand for a full run of it again... Let me know if you're one of those who wants one if we were to do another production run!
Everyday Astronaut tweet mediaEveryday Astronaut tweet mediaEveryday Astronaut tweet mediaEveryday Astronaut tweet media
English
42
9
431
38.4K
Gael Ollivier
Gael Ollivier@gaelollivierdev·
@Ike_Saul @arijoe19 As a software engineer I keep asking myself “at this rate of progress, what are the tasks the AI *won’t* be able to do in 6 months, 1 year…? And the scaring thing is I can’t think of any. And I can’t think of many jobs/industries that won’t be impacted in the next 2-5 years
English
0
0
0
49
Gael Ollivier
Gael Ollivier@gaelollivierdev·
@Ike_Saul @arijoe19 The AI has to try things, make hypothesis, make mistakes, and improve from its learnings. And that’s what software engineers are witnessing today: the AI is really doing all these things in a way that didn’t seem possible even 6 months ago…
English
1
0
0
51
Isaac Saul
Isaac Saul@Ike_Saul·
This piece is interesting. I also think it's deeply flawed. One thing I've noticed, and h/t to @arijoe19 for articulating it so well, is that computer code is a really structured language, and software is a defined problem space with a lot of defined patterns, so software people tend to think everything is a pattern and AI being really good at their job makes them overestimate how well it can do everything else. The truth is there is a lot more disorder, unpredictability, and humanness in so much of our lives — and our work — that I don't think AI applications will always (or even often?) be able to account for. Matt, for instance, lists journalism as a job in trouble thanks to AI (not that our industry needs more trouble). And it's true that AI can read documents fast and do incredible research and even write clean copy and edit -- it will probably eliminate or reduce the need for some jobs! But you know what it can't do? It can't work a source over for years on end. It can't / doesn't / won't bear witness to live events. It reminds me of the famous Good Will Hunting scene, where Robin Williams is chastising Matt Damon about being such a smart ass but not being able to describe what the Sistine Chapel smells like. Damon is the AI. I say this as someone who has experimented a ton with the latest versions of ChatGPT Matt is writing about here. I can feed it limitless writing of mine from my archives and then have it write a take about a new current events story; I've tried, actually, because if it were good it would save me hours of work every day. But it is *always* useless. Not sometimes; always. Why? Because the AI still can't predict when certain emotional elements of a story drive me away from a previously held position; because it doesn't know what happened to me that week, or what stories I've read about the topic at hand, or an experience my grandmother had that my family always talked about that informs my view on, say, antisemitism or Israel. It just predicts where I'd land on an issue based on what I've written before, which is actually not a great way to understand humans who are always moving in new and different directions. It just doesn't know. People think humans are finite numbers of neurons and processes and thoughts and learning but I think that is wrong — we are all constantly changing every day, every second, thanks to new inputs and new experiences. So yes, I buy that AI will be able to read documents better than your typical lawyer. But can it build a relationship with a client? Or look at a jury and guess what argument might move them to "guilty"? Or know when to cross the lines with a judge or when to step back? I don't really think so. And those limits, to me, are so under-discussed in this dialogue that it kind of discredits everything else.
Matt Shumer@mattshumer_

x.com/i/article/2021…

English
124
231
1.7K
386K
Gael Ollivier
Gael Ollivier@gaelollivierdev·
@joebarnard I want to know if the scammers used memes though, otherwise it would clearly have no resemblance to the original
English
0
0
0
78
Joe Barnard 🚀
Joe Barnard 🚀@joebarnard·
I will NEVER direct message you asking for money - if you ever get a message from an account that looks like me and wants you to donate or send money, block and report them. Thanks!
English
23
7
322
12.5K
Gael Ollivier
Gael Ollivier@gaelollivierdev·
@Erdayastronaut These companies are optimizing for engagement and ad revenue. The majority of people don’t notice or even care that they’re just doom scrolling through AI slop. So the algorithms keep pushing the brain rot content…
English
0
0
2
89
Everyday Astronaut
Everyday Astronaut@Erdayastronaut·
Like what’s the end goal? AI has the ability to create infinitely more content than humans and flood the internet with 99.999999% of content, burying anything that is actually meaningful and connecting, completely ruining the whole point of social media. Make it make sense
English
40
16
356
20.2K
Everyday Astronaut
Everyday Astronaut@Erdayastronaut·
Does anyone else feel like social media companies are creating their own demise by heavily investing in AI content? Each time I login I feel more disconnected w/ my friends, creators, events & brands that I used to go to connect with. Advertisers $ must be less effective, right?
English
164
43
1.6K
67.1K
Gael Ollivier
Gael Ollivier@gaelollivierdev·
@DJSnM I’m at a team onsite in SSF and thought “oh I didn’t know SF had a blimp”. Sounds like it’s not actually that common!
English
0
0
0
173
Scott Manley
Scott Manley@DJSnM·
I saw the airship with my own eyes at last:
Scott Manley tweet mediaScott Manley tweet mediaScott Manley tweet media
English
14
8
307
22.9K
NSF - NASASpaceflight.com
NSF - NASASpaceflight.com@NASASpaceflight·
LAUNCH! Starship Flight 11 sets sail with Booster 15-2 and Ship 38. The final Block 2 (V2) Starship and the final launch from Pad 1 in its current configuration. Live on X and YT: youtube.com/watch?v=7bcpnn…
YouTube video
YouTube
English
14
109
1K
55.5K
Gael Ollivier
Gael Ollivier@gaelollivierdev·
@DJSnM I was standing there 1h before the crash, we saw the parking closed and the filming car, then 3 copters showed up doing formation flying and then started doing low passes and landings on the parking lot
English
1
0
4
430
Scott Manley
Scott Manley@DJSnM·
I guess the Helicopter that crashed was supposed to be landing at this event:
Scott Manley tweet media
English
1
2
110
20K
Gael Ollivier
Gael Ollivier@gaelollivierdev·
@DJSnM They were filming and probably rehearsing for the “cars and copters” event tomorrow… I was there an hour earlier and saw other copters fly right above us and land on the parking lot. Kinda shocked to see this on my way home…
English
2
0
22
1.7K
Gael Ollivier
Gael Ollivier@gaelollivierdev·
@DJSnM There were multiple copters making low passes and landing on this parking lot, apparently preparing for a cars & copters show tomorrow. I was right there a few hours before with my wife and daughter. This is so scary…
English
0
0
3
1.4K
Scott Manley
Scott Manley@DJSnM·
I won’t be surprised if they got something like a persistent chip light warning them of imminent gearbox failure and decided to land ASAP
English
10
4
123
18.2K
Gael Ollivier
Gael Ollivier@gaelollivierdev·
@DJSnM I’m a long time follower and suddenly your videos became harder to find, if I don’t click it as soon as it’s out (I usually take my time) I might need to be looking for it explicitly. I definitely noticed a change…
English
0
0
0
335
Scott Manley
Scott Manley@DJSnM·
Whatever changed had a much bigger impact on the new and casual viewers. Maybe every other creator just got better and my thumbnails suck.
Scott Manley tweet media
English
32
2
176
22.5K
Scott Manley
Scott Manley@DJSnM·
Since YouTube introduced A/B testing for titles I’ve lost half my views, almost no traction from non-subscribers any more. I’ve no idea what I should be doing to fix this.
Scott Manley tweet media
English
308
34
2.1K
202.9K
Gael Ollivier
Gael Ollivier@gaelollivierdev·
@DJSnM The link says « milliseconds », I was surprised when I read « seconds » as you would think the stages would be already far enough that a first stage failure wouldn’t break the 2nd stage at that point?
English
0
0
1
464
Scott Manley
Scott Manley@DJSnM·
So what I'm reading is the booster failed seconds after stage separation, and that failure was energetic enough to destroy the nozzle extension on the second stage. LOL, ignore the rest, I'm completely misunderstanding the reference to flow separation. Remember, I'm not a rocket scientist I only play one on the internet.
Firefly Aerospace@FireflySpace

Firefly Aerospace received Federal Aviation Administration (FAA) clearance to resume Alpha rocket launches. fireflyspace.com/missions/alpha…

English
9
1
219
33.5K
Gael Ollivier
Gael Ollivier@gaelollivierdev·
@Ike_Saul It’s a typical authoritarian “us vs them” move. I represent the truth, and if you disagree, you’re not one of us, but one of those evil “them” enemies
English
0
0
0
101
Isaac Saul
Isaac Saul@Ike_Saul·
1/ Trump explicitly says he no longer wants the support of anyone who his base who is stuck on the Epstein files. Pretty wild -- not sure I've ever seen a politician just give the middle finger to his core supporters like this. But...
Isaac Saul tweet media
English
24
41
393
28.1K
Gael Ollivier
Gael Ollivier@gaelollivierdev·
@bsansouci It’s pretty common to see ads that aren’t labeled as such on the “Following” tab, unfortunately…
English
0
0
1
12
Ben
Ben@bsansouci·
This is not labelled as an ad, I don’t follow them, I’m on the following tab, and nobody I follow retweeted this. Why am I seeing this spam
Ben tweet media
English
1
0
1
185
Gael Ollivier
Gael Ollivier@gaelollivierdev·
@DJSnM Kinda cool that is basically a coaxial tank that also serves as a downcomer tube 🤯
English
0
1
1
274