ttul

35 ms·

631.
by ttul 3y ago
This would make for a nice ComfyUI custom node.
632.
by ttul 3y ago
Gotta have those Mac laptops for the wunderkind of AI.
633.
by ttul 3y ago
Well, minutes after posting this, I got an invite.
634.
by ttul 3y ago
Yeah, I have been waiting for six months. And I have a real company with a real use case. I guess demand is off the charts.
635.
by ttul 3y ago
Many logs in the Salish Sea are actually enormous tree trunks complete with root systems…
636.
by ttul 3y ago
It always confused me that we would sit in these large offices together, chatting with each other on Slack. It was like the idea of working from an office had greatly outlived its utility following the development of the internet, yet nobod
637.
by ttul 3y ago
“I hope this email finds you well…”
638.
by ttul 3y ago
When I was taking Jeremy Howard’s course last fall, the breakthrough in SD was going from 1000 steps to 50 steps via classifier-free guidance, which is this neat hack where you run inference with your conditioning and without and then mix t
639.
by ttul 3y ago
To me, the crazy thing about LoRA is they work perfectly well adapting models checkpoints that were themselves derived from the base model on which the LoRA was trained. So you can take the LCM LoRA for SD1.5 and it works perfectly well on,
640.
by ttul 3y ago
I’m pretty sure that any legislation making ransomsware payments unlawful could carve out exceptions to deal with national security matters.
641.
by ttul 3y ago
If you use a tool like ComfyUI, you can use LCM-LoRa to generate fast but not perfect outputs and then refine them with the old school sampler. I’ve been playing around and have found the quality of LCM to be excellent when used in combinat
642.
by ttul 3y ago
If we want to stop ransomeware, then government should make it illegal to pay ransoms. If CFOs faced jail, then companies hit with attacks would never pay. After a few such failures, gangs would focus on countries that don’t prohibit paymen
643.
by ttul 3y ago
Yeah, indeed. I know LCMs are way faster than diffusion models, but surely this demo is running on a beastly GPU...
644.
by ttul 3y ago
The progress here is remarkable. A year ago, we didn’t even have ChatGPT. LLM completions were cool but so hard to use and definitely there was nothing accessible to non-nerds.
645.
by ttul 3y ago
I recall dimly a period of months during engineering school when I would have been able to parse those symbols and perhaps make a joke about something in the lunch room. Those days are long behind me.
646.
by ttul 3y ago
Does anyone know whether he will be re-bailed to his palatial parents' home in Palo Alto now that witness tampering is off the table?
647.
by ttul 3y ago
The Kemano generating station in British Columbia was built in the 1950s to supply power to an aluminum smelter. They hollowed out a mountain to drop a river through turbines. It’s an incredible marvel of engineering. https://en.
648.
by ttul 3y ago
Spy satellites are rumoured to use inflatable stealth shields to deflect radar and optical sensors so that they can hide from adversaries: https://patents.google.com/patent/US5345238 "An inflatable shield for supp
649.
by ttul 3y ago
Partly they do this to keep their ML researchers entertained after the glitz of their $1M compensation wears off. And partly they do this to develop products for their customers. Although they partnered with Anthropic, whose arguably SOTA m
650.
by ttul 3y ago
It's all history, but it does sting somewhat to realize that so much VC went to the undeserving founder of WeWork, enriching him enormously while the eventual funding all bled away to landlords and banks - instead of other more deservi
651.
by ttul 3y ago
TL;DR With or without the PM title or experience, hire smart people.
652.
by ttul 3y ago
There may be some interest in having access to Google’s machine learning hardware as well. What we don’t know publicly is the extent of any deals these companies have to share technology and infrastructure. Google is one of the only hypersc
653.
by ttul 3y ago
They are buying upside and they are funding a competitor to OpenAI and Microsoft. The enemy of my enemy is my friend. There are only so many groups in AI that have the chops to take on OpenAI and Anthropic is one of them - if not the most p
654.
by ttul 3y ago
I know.
655.
by ttul 3y ago
That depends on whether the training data contained languages other than English.
656.
by ttul 3y ago
This is super cool. Writing code to drop into the JavaScript console lets you do insane things. I’ve found great success using ChatGPT to help me write the code, which I then just cut and paste into the console. Asking it to “make it all ru
657.
by ttul 3y ago
I suppose there is no reason you can’t profile your code and have a tool insert these hints based on actual statistics from execution.
658.
by ttul 3y ago
“Basically a taxi service,” except that Uber spans hundreds of cities, coordinates millions of drivers - none of whom work on a fixed schedule - and its only interface with customers is an app that has to be fast, accurate, and reliable at
659.
by ttul 3y ago
It may not be. It’s hard to tell. The semiconductor business is extremely hard to understand unless you’re steeped in it.
660.
by ttul 3y ago
This worked for me: https://chat.openai.com/share/6e2bbc7c-0a87-4885-8dbe-2b0900... Basically you ask for the tokens and if they aren’t provided, then ask GPT to generate a Python data structure containing the tokens.

More ›