Justin Beausoleil (@jusbeau) header
Justin Beausoleil (@jusbeau) profile picture

Digital marketing director and early adopter. Deep down the AI rabbit hole, exploring what I can create, automate and improve for myself and my clients.

Barbados
Joined October 2008
Photo by Justin Beausoleil (@jusbeau)Photo by Justin Beausoleil (@jusbeau)Photo by Justin Beausoleil (@jusbeau)Photo by Justin Beausoleil (@jusbeau)Photo by Justin Beausoleil (@jusbeau)Photo by Justin Beausoleil (@jusbeau)Photo by Justin Beausoleil (@jusbeau)Photo by Justin Beausoleil (@jusbeau)Photo by Justin Beausoleil (@jusbeau)Photo by Justin Beausoleil (@jusbeau)
Justin Beausoleil retweeted
The SpaceX engineers who caught the rocket don't write LinkedIn thought pieces about leadership.
718
6,840
386
75,352
1,853,818
How do I opt out of these translated posts appearing in my feed?!
17
Justin Beausoleil retweeted
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.
1,820
6,428
2,201
41,231
6,737,120
Me knowing you can’t take a screenshot of a post directly from the copy link feature but still trying anyway.
27
Never deleting this app
🌊💙 WE WERE ALWAYS BLUE
1
96
Justin Beausoleil retweeted
So we are finally opening our new app that we use internally for a few weeks. It is a Linear / Notion / Slack / Discord / Mailing all in one app that you can self-host on a few MB of RAM and own all the data. orbit.so There is also a cloud/hosted version with 7 days of free trial so you can test things out. After that it is just $10/month/workspace. No per seat or similar bs. $10 for your whole team. I will go broke for sure, but you won't :D (I have to at least get back the domain price somehow to get on 0) Same philosophy as any coolLabs apps, no restrictions, no pay-to-feature. Lets Orbit together with your team haha!
173
84
26
1,749
88,272
Justin Beausoleil retweeted
306
6,000
621
63,963
1,492,442
So why will it work this time around?
43
Justin Beausoleil retweeted
How you do anything is how you do everything.
Saying please and thanks to your agent feels like a waste of tokens… but our brain can’t really tell the difference between talking to your agent and talking to a human. So if you command your AI all day without being polite, it'll transfer to how you talk to colleagues, your wife, your kids and so on. Don't talk politely for the AI's sake. Do it for your own
100
89
25
2,084
96,686
Justin Beausoleil retweeted
Steve Jobs on Touchscreen MacBook: “It doesn’t work, it’s ergonomically terrible”
252
829
284
6,712
655,969
Nobody asked for this. They are running out of ideas.
We brought Touch to the Mac! From the all-new iPad mini to the 14-inch MacBook Pro and iMac... and now MacBook Pro with PROLED XDR Display and Touch, creators have new tools to do their best work yet.
40
Too many arc alternatives now. I can’t keep up.
Introducing Sailfish, the open-source alternative to Arc. I didn’t like Dia and wanted to keep using Arc long-term, so @josephtso914 and I built it ourselves + open-sourced it. iOS app as well. You can also import your settings! Check it out here: sail.fish
45
90% accidental clicks from Instagram.
JUST IN: Mark Zuckerberg says Threads has surpassed X in monthly active users
1
1
38
Been using @hark_labs for a couple weeks now. Most useful and proactive agent I’ve tried so far, plus awesome UI/UX✨ Really impressive computer use but more integrations would be better. Looking forward to seeing where this goes.
1
92
Justin Beausoleil retweeted
If you have an existing Cursor, Grok, and or 𝕏 subscription, here's how to unify/combine all your accounts into the new 𝕏 Subscription: When you switch from your current plan, eligible monthly and annual plans are cancelled, and you get a prorated refund for unused time. The refund goes back to your original payment method separately from your checkout total, so your new plan is charged at full price. If you pay for Cursor through the Apple App Store, switching doesn't cancel that subscription, and 𝕏 can't refund it. Cancel it in your iPhone or iPad Settings so Apple stops billing you, and request any refund from Apple. Switching is optional. If you don’t switch, your current plan continues unless you change or cancel it. Existing Cursor, Grok, or X subscribers can switch to The X Subscription and unify eligible accounts at cursor.com/unify (Cursor), accounts.x.ai/unify (Grok), or on x.com (X) via the in-product banner. If you already pay for Cursor, Grok, or X, switch on the web at one of these links. For now, the mobile apps send you to the web to switch. If you don't pay for any of them yet, you can set up a new account and subscribe to The X Subscription in the Grok, X, Grok Bot, or Cursor mobile apps, if desired. These purchases are billed through the App Store or Google Play Store. Switching unifies eligible Cursor and Grok accounts into one account with one workspace, one plan, and one bill. Adding a personal X account gives you X benefits under The X Subscription. A completed unification is permanent. It can’t be undone, and Support can’t undo it for you. After you unify, you manage your plan, billing, and settings at grok.com/dashboard. When you switch, your old Cursor, Grok, and X Premium subscriptions are cancelled automatically, so you don’t need to cancel them yourself. Any refund for unused time is sent separately. If your X, Grok, and Cursor accounts use different email addresses, they don’t need to match. 𝕏 will connect the X account you sign in to, not an account that shares your email. When you switch, sign in to the exact Cursor, Grok, and X accounts you want to combine, and check each one before you confirm. Keep work and personal accounts separate unless you mean to combine them. If the wrong account appears, stop before checkout and contact Support. Learn more: docs.x.ai/x-subscription/uni…
SpaceX has just officially introduced a new "𝕏 subscription" that combines Grok, 𝕏 and Cursor into one plan. Here are the tiers & info: • Plus: $8/month • Premium: $30/month • Super: $100/month • Ultra: $200/month Plus: $8/month • Cursor access • Grok Chat, Voice, Imagine & Build • Grok API • Shared usage pool across products • Premium checkmark • Fewer 𝕏 timeline ads • Creator monetization • Edit posts • Longer posts & video uploads on 𝕏 • Articles & Media Studio • Advanced Analytics • Higher posting & DM limits • Bookmark folders • Download videos Premium: $30/month • Includes everything in Plus, and: • 4x more usage compared to Plus • Grok Bot • Latest Grok models • Faster Chat responses • Al images & videos • Ad-free scrolling on most surfaces on 𝕏 • Preferred @handle • Higher reply boost Super: $100/month • Includes everything in Premium, and: • 15x more usage compared to Plus • Priority model access at peak times • Faster Chat responses & generations Ultra: $200/month • Includes everything in Super, and: • 50x more usage compared to Plus • Earliest access to all products & models • Fastest Chat responses & generations Compare features: x.ai/pricing
239
152
50
1,980
204,958
Justin Beausoleil retweeted
Are minimalist logos a trend? What do you think? Let’s talk about it!
34
38
6
1,108
32,293
Justin Beausoleil retweeted
Hark has exploded over the last 48 hours, and all the user growth caused some infra issues today. fwiw launch was far beyond my expectations I'm personally on it with the team; we're actively working to stabilize
86
19
5
699
83,443