I didn't install OpenClaw explicitly ๐ฏ๐ฒ๐ฐ๐ฎ๐๐๐ฒ ๐ผ๐ณ ๐๐ต๐ฒ ๐ผ๐ฏ๐๐ถ๐ผ๐๐ ๐ฐ๐๐ฏ๐ฒ๐ฟ๐๐ฒ๐ฐ๐๐ฟ๐ถ๐๐ ๐ถ๐๐๐๐ฒ๐. So, I can see this exact scenario playing out:
This acquihire means your OpenClaw/ClawdBot will start to *prefer* OpenAI models and Sam Altman will productize a safer, normie-friendly version of the bot as a subscription service in ChatGPT:
I teach auto shop at a small high school. We work on students cars, teachers cars, students parents cars and some community people cars. We only charge for parts and not labor, so we saved some people a lot of money last school year. This last school year we did 126 oil changes, 68 brake jobs, 85 alignments, 4 steering racks, 22 tune ups, 32 struts, 20 shock absorbers, 4 transfer cases, mounted and balanced 82 new tires, 4 timing chains, 15 valve cover gaskets, 14 thermostats, 4 radiators, 12 in tank fuel pumps, 8 EVAP canisters, 6 exhaust manifolds, 4 mufflers, 15 AC repairs including evacuate and recharge, 8 alternators, 22 batteries, 9 starters and so much more! Proud of those students I am!
How much money have you spent on art over the last decade? I consider myself lucky to have been able to support so many artists over the years, I kind of wish PFP summer would come back so I could flip jpegs and continue to support artists the way I did before...
Even with GPUs at insane prices, CTOs are tempted to "๐ฃ๐ถ๐บ ๐ฐ๐ฏ๐ค๐ฆ, ๐ค๐ณ๐บ ๐ฐ๐ฏ๐ค๐ฆ" to own an AI GPU cloud.
But there ๐๐๐ hidden costs. Power, cooling, and security add up to 40% more than the purchase price of just the AI chips.
Read more: https://t.co/eBX5Qr2QuS
Another reality of these compute-based satellites is that they don't need to be deployed in mass quantity to get useful compute. Launch one, load its model, add it to the inference cluster and its crunching AI.
This is very different than the critical mass problem Starlink experienced in the early days. You needed a lot to build a 'network' for a viable service.
I just sat through Morgan Stanley's ๐ ๐๐ฐ๐ฏ๐ท๐ฆ๐ณ๐ด๐ข๐ต๐ช๐ฐ๐ฏ ๐ธ๐ช๐ต๐ฉ ๐๐ฑ๐ข๐ค๐ฆ๐ webinar which was promoted as a must-attend event featuring C-suite SpaceX execs talking about the nascent SpaceX IPO.
I learned nothing about the IPO and it was a waste of time.
Lots of AI announcements at Apple's WWDC event, but the biggest disappointment was no new "Apple Home" devices to run these consumer-based AI features on.
OK, the AI features are almost wholly aimed at consumers who don't know what an OpenClaw or Hermes is. Or have tried the new @TownAI, @wabi, or @interaction's Poke.
It touches very lightly on things that could take away people's privacy. Like looking at your screen. Others, like @FarzaTV's Clicky or @cluely dive deeply into that side of the pool that Apple is afraid of popping in.
Very few words about glasses, or Apple Vision Pro. This was all about a rebuilt Siri. Which is great for everyday users (most of my family and friends who aren't in AI world will be thrilled to get them).
Watching it I realized just how far ahead of the world X's AI community is. If you want to get ahead, follow everyone here: https://t.co/9eRY65x3IQ
And I'll have my AI write up a report from all that at https://t.co/8L5xphk0qQ
Prices high from a lack of GPU chips? It's also a dearth of HBM and power that constrain 2026 AI compute.
What's the winning strategy for CTOs? We recommend finding partners with neocloud capacity... GPUSeeker can help!
Learn more: https://t.co/xNe0GhjXDE
Concerning that Tesla promotes pro-shot footage of FSD flawlessly navigating impossibly-narrow Chinese mountain roads but I can't find video of even a private Tesla owner using FSD on Maui's Road to Hana.
@Wingmaster85@dallasteslaclub As a known FSD enjoyer, I will reiterate my view that people think they want to/like to driveโฆ but they really donโt. That is because they havenโt experienced FSD for themselves. RoboTaxi + World Cup fixes that.
The longest Iโve lasted this year was 3 days without FSD.
Undeniable that we're experiencing a GPU shortage. But it's less of a "no bread on the shelves" and more of a "loaf of bread costs $1,099" kind of thing...
Whether it's your e-commerce site or streaming video app, AI product recommendation engines requires precise VRAM planning. Do you know if your infrastructure can handle the load? Read our guide to the GPU memory requirements for popular AI models. https://t.co/KpzlLBbHr8
Finland is a great place for Nebius because about 38% of their nation's power is from nuclear, 17% from hydroelectric, and they have natural year-round cooling (lowers overall electricity consumption).
Meta is the only major hyperscaler that does NOT resell cloud compute today. Now the company is signaling that business pivot's "definitely on the table," but only if they have overcapacity of GPUs. I bet they find a use for their hardware... https://t.co/tavXUMTnn1
When I interview candidates, I always ask about their willingness to work in the factory.
One said he was a researcher, not a factory worker, so I asked how he measured part-to-part variance for a part on his resume. He could not explain. All he had done was training without considering the data distribution from different manufacturing processes.
In our realm, you do whatever it takes to understand a problem. If it involves chemistry, you learn chemistry. If it involves aerodynamics, you learn aerodynamics. There is nothing you cannot learn or do. Thus, the distinction between research and engineering is extremely blurred.