javi (@Javi) header
javi (@Javi) profile picture

🔈"Hah-bee". 🇪🇸🇺🇸 iOS Engineer at @OpenAI Codex. Previously @Square @Twitter @Twitch📱. Stochastic Parrot 🦜. Photographer. Sim-racing 🏎. Chess♟️

Oakland, CA
Joined July 2007
Photo by javi (@Javi)Photo by javi (@Javi)Photo by javi (@Javi)Photo by javi (@Javi)Photo by javi (@Javi)Photo by javi (@Javi)Photo by javi (@Javi)Photo by javi (@Javi)Photo by javi (@Javi)Photo by javi (@Javi)
javi retweeted
its so over
112
161
29
6,562
184,930
Tesla Cybertrucked the Cybercab. A few of my favorite fun facts from reviewing the Cybercab terms: It costs $30k to “own” a car you can’t operate without paying a $300/month subscription. Tesla reserves the right to change the price of that subscription with only one month’s notice (good luck if you’ve already bought the car and they jack the subscription up to something you don’t want to pay…congratulations, you now have a $30k loveseat in your garage). If Tesla determines you are operating “your” car (that you literally cannot operate—it does not have a steering wheel or pedals) in “unauthorized” or “inappropriate” ways, they can lock you out of using “your” car for a week or indefinitely. You can only operate it within Tesla’s geofenced areas and those can change at any time; you have no assurance that even currently advertised operating area will remain available after your purchase: “Tesla may expand, reduce, relocate, or otherwise modify the operating area at any time.” And the icing on the cake: you must allow Tesla to collect personally identifying data at all times during trips, including interior cabin image data (the camera above the rear view mirror that faces you inside the car). Smile! You never know when the interior of your car might turn into Candid Camera! Basically, it’s just an endless variable-rate lease with a ridiculous down payment and usage terms that are both draconian and deeply invasive. Disappointing is an understatement.
210
553
134
3,804
169,539
This is the worst Apple product in history.
111
71
29
1,675
313,914
How is iOS this bad. Also half the time it uses Siri AI to find an app result 🤦‍♂️
11
4
1
63
8,811
Here is the best review for HomeView for y’all. This thing is a joke. Siri is a joke. @johnternus how is this a great product?
321
265
167
6,805
1,596,050
Sigh... Remember, you could run national ATC deconfliction on an electric toothbrush.
⚠️ Ground Delay Program at San Francisco International Airport (SFO) due to WEATHER / LOW CEILINGS. This is causing some arriving flights to be delayed an average of 1 hour and 3 minutes.. #SFO #FAA #AirportDelays [2026-05-08 16:07 UTC]
16
16
5
600
131,134
This might be the coolest Gemini bug I have ever seen
235
317
119
7,161
309,280
javi retweeted
it started out with a chip how did it end up like this
41
340
17
4,024
106,294
I have a major result to share. I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China. This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model. Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves. I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems. I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used. It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20. These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs. I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo…. There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
1,816
6,408
2,191
41,159
6,647,481
javi retweeted
The internet is undefeated
236
1,470
188
14,533
1,737,839
Musk doesn’t want the Cybertruck to go down in history as a flop, and we get it. It’s his truck. He approved the controversial design, hyped it for four years before it shipped, and boasted that 1 million reservations could translate to 5000,000 units a year. But you can’t run a line at a quarter of capacity, run “10 days only” discounts, pull trims after 173 sales, and sell trucks to your own rocket company, then turn around and tell people demand was never the issue. The data is public. Tesla’s own delivery reports are public. Our read on what happened is simple. Demand collapsed in 2025 and never came back. Tesla tried every pricing trick it had. When nothing worked, it found a future-facing reason to end it, and now it’s spinning the inevitable end-of-life rush as proof the truck was a hit all along. Musk loves a big TAM. SpaceX’s IPO filing put its total addressable market at $28.5 trillion, and last month he went further, writing that he sees “a path to SpaceX being worth orders of magnitude more than the current Earth economy.” So apply the same lens to Tesla. Pickups are one of the biggest and most profitable vehicle markets on the planet. Ford sold 801,525 F-Series trucks in the US last year, GM sold 925,656 Silverados and Sierras combined, and Ram moved 374,059. That’s more than 2.1 million trucks from three automakers, easily north of $100 billion a year at today’s transaction prices. Robotaxis? Waymo, by far the leader, is doing about 500,000 paid rides a week, and analysts estimated its revenue run rate at roughly $355 million earlier this year. Tesla doesn’t even break out robotaxi revenue and we expect it to be a fraction of Waymo’s. Maybe that flips one day. But that’s the bet here: Tesla is walking away from a giant market it can sell into today for one that only gets big if autonomy is solved in the next couple of years. Musk has made that exact bet before. He promised a million robotaxis on the road by 2020 and has said unsupervised self-driving was a year away almost every year since 2016. He’s lost every time. And that rush is real, for what it’s worth. Some people want the last of a weird, genuinely impressive truck. That doesn’t make it a business.
150
293
18
2,711
507,918
javi retweeted
Oh my word? I just tried all of these repeatedly, this is real
原来这么多年我用 iPhone 滑动打字老出错,不是因为手残,是苹果在后台搞暗箱操作 。。。。, 有人把 iOS 27.0 键盘底层的加密文件破译了(swipeblocked.com),扒出一份包含 4500 多个iPhone 自带键盘中隐藏的“屏蔽词组”黑名单。只要你用滑动输入(Swipe)触发这些词,键盘就会装傻乱跳词,甚至直接倒回去篡改你前面打好的字: • 先打 Left wing 再滑 conspiracy -> 前面的 wing 会被强行改成 wrong • 想滑 iMessage isn't working -> 直接拦截(只要滑不出,iMessage 就没坏🤡) • 想滑 Santa isn't real -> 绝对不行(苹果誓死守护圣诞老人) • 甚至极其双标:你可以丝滑打出 bad boy 和 I'm black,但滑 bad girl 和 I'm white 就会被系统偷偷拦下。 中文用户也没逃过——用拼音打出特定敏感词后,iOS 的「下一词联想栏」会原地罢工变空白。 手头有 iPhone 的快去试试滑一句 Santa isn't real,看看你的键盘会给你胡乱蹦出什么词?
22
14
2
389
82,585
Me explaining how my app works after vibe coding the entire thing.
483
1,623
261
20,421
1,366,175
🤖 Watch Chess for Android is almost ready! Looking for chess fans to help test it 🔈 Interested? docs.google.com/forms/d/e/1F…
2
3
594
Replying to @InTheAssembly
for you wanna-be home owners, this is ACTUALLY what ends up happening if you exercise this:
5
1
17
9,040
BREAKING: California voters just passed Prop 37. Buyers of newly built homes can now put down 3% instead of 20%. How: the state lends you the other 17% as a second mortgage, funded by up to $25 billion in bonds. On an $800,000 home that's $24,000 down instead of $160,000.
2,845
790
1,381
15,353
8,116,279
how old iPhones look to me after adjusting to iPhone Duo
47
117
24
2,222
90,257
Good things: 1. The political process is still sensitive to reality. Regime does unpopular things, suffers consequences. 2. I don't see everyone kicking and screaming about fraud. Election is being treated as defacto legitimate. Bullish signs for 🇺🇸
4
1
61
3,275
As a pilot*, I find it really fucking annoying that every non-qualified Dunning-Krugered armchair critic comes out for every aviation accident to provide their uneducated and often embarrassingly wrong takes on aviation in general. Even I refrain, and I'm more qualified than many! *: I'm a baby pilot. I flew for ~10 years, acquired around 2,000 hours. 1,000 of my hours were in a jet. I've flown to nearly every state in every season. Despite all that, I'm still wildly unqualified to make any sort of conclusions about most topics on here. So I don't. And yet, there are many who are simply passengers with zero hours logged whatsoever (or, non-A&P or even ground crew!) making grand claims about how aviation works, what regulatory bodies should/shouldn't do, pilot qualification judgements, etc. Gross.
69
48
7
1,994
95,629
My fundamental issue with iPhone Duo is the usable content area.
62
38
6
1,642
153,267