Co-Founder/COO of 3D body tracker @shapescale. 💪 Data & fitness tech geek, dad and husband. 🏃👊🔥

San Francisco Bay Area
Joined February 2009
Include

Only show posts containing:

Exclude

Hide posts containing:

Time range
-
Minimum likes
Replying to @niklaslenz_ai
I'm using DeepSeek V4.1 Flash iQ3 3.0bpw with mlx-serve. Bigger quant, and up to ~100 tok/s, though at ~400k ctx it's more like 60-70 tok/s.
4
73
This is such a dumb take. As with any diet… You still need to work out, you still need to eat protein, if you want to keep your muscle.
Gabriel Iglesias was on Ozempic for about five months. His description is the most honest one I’ve heard. It worked almost too well. He’d look at food and feel nothing, not disgust, just “I’m good.” The weight came off. But he watched his shoulders disappear with it. A strong guy, losing the strength he was proud of. Add the constant nausea, and he quit. The moment he stopped, 20 pounds came back almost immediately. And here’s the twist: years earlier, he says Ozempic offered him a million-dollar deal to be the face of the brand. He turned it down. That’s the part nobody puts in the commercial. The drug turns off your hunger. It doesn’t protect your muscle, and it doesn’t fix whatever was driving the eating in the first place.
1
66
Replying to @alexandr_wang
Should have differentiated the model name better from regular Muse Wonder.
377
Replying to @trevin
It’s like on one hand they need to cut usage but then on the other its like resets like no tomorrow. Maybe that was always the plan? 😅
11
Replying to @DanielGri @pidotdev
Interesting. Reminds me a bit of @kunchenguid ‘s Firstmate but just for individual projects if I understand correctly.
162
Replying to @vr8vr8
Great work. I gotta wonder though other than having more total KV cache and one model endpoint, what's really the advantage of running tp=2 vs running both sparks on tp=1, assuming c>4? Seems like I'd be better off running them separately.
2
37
Yes a trifold makes much more sense form factor wise. That’s what I’d hold out as well. Otherwise the compromises just aren’t worth it.
1
36
Wouldn't be surprised. You can almost not browse any accounts with default Muse browser these days.
Prediction: you’re all going to be using @AsideAI with your agents in a year or so
128
Replying to @thsottiaux
Too early 🥲
2,669
Replying to @majewskizby
That GLM 5.3 prefill looks a little low. You sure about that?
1
65
This is pretty cool.
Plus up your furniture with 3D-printed accessories using Gemini 4 Argon. Upload a pic with rough measurements, and use Canvas in Gemini to build an interactive 3D CAD studio where you can explore styles, tweak live sliders, and export printable STLs to upgrade your space.

25
Replying to @mariagorskikh
lol that's pretty clever
1
22
... and resets every other day! 🫠 Otherwise I'd probably get more out of Opus/Sonnet/Haiku.
Token spend has moved from Anthropic to OpenAI because - Luna is the king of small closed models - Sol 6.1 is an excellent every day driver for non-coders - Astra for coding is amazing, if you can afford it. Less dense than Fable Overall OpenAI has a very solid model family
50
Replying to @simonw
But then how much revenue goes a loss due to it? 🤔
1
2
374
Replying to @TJeparskis
You'd hope they had 28 things planned out already. But knowing OpenAI it's probably a true challenge and they're grinding away on this. Maybe not the worst idea?
61
Martin Kessler retweeted
A hundred years ago, if your kid got diabetes, the doctor's plan was to starve them. 400 calories a day. Some weeks, nothing at all. It sounds insane but there was a grim logic behind it. Type 1 diabetes killed children within months back then, and less food meant less sugar in the blood, which slowed the dying. Kids on the diet stopped dying of diabetes and died of starvation instead, just slower. Parents signed up anyway because one extra year with their child was the best medicine could offer. Elizabeth Hughes was the daughter of the US Secretary of State, so she got the best care money could buy. That care took her from 75 pounds at diagnosis down to 45 pounds by age 14. When her mother brought her to Toronto in August 1922, she could barely walk. The treatment waiting there almost never got made. Frederick Banting was a small-town surgeon whose practice was failing so badly he was nearly broke. He talked a university into lending him a lab for one summer, and with a 21-year-old student assisting, isolated the pancreatic hormone researchers had chased for 30 years. Five weeks after her first shot, Elizabeth was eating 2,500 calories a day with a pint of cream poured on top. She gained 60 pounds in four and a half months. She went to Barnard, raised a family, lived to 73, and kept the diabetes secret from almost everyone she knew. Banting refused to put his name on the patent. He'd sworn the Hippocratic oath and believed a doctor had no business profiting from something like this. His co-discoverers sold it to the University of Toronto for a dollar. "Insulin does not belong to me, it belongs to the world." 24 months from first injection to the Nobel Prize, and every diabetic kid on earth went from a death sentence to a normal life for the price of one dollar.
In 1922, this little girl was wasting away from type 1 diabetes. Four months after doctors gave her a new experimental drug called insulin, she looked like a different child.
99
2,007
77
23,945
2,154,198
Well you could run GLM 5.3 on 4 sparks but it won't be very fast. And that's still easily $20-25K.
33