syntaxing

5 ms·

31.
by syntaxing 1mo ago
Overkill is the point. Same thing for iPads and iPhones. It gives a luxury feeling. That and Chinese manufacturing have made milling a very cost effective manufacturing method.
32.
by syntaxing 1mo ago
iPhones are iPads have a milled CNC “unibody” as well… Being able to dictate where to have more thickness helps a ton on the luxury feel. Most of the chips are recycled anyway into new billets.
33.
by syntaxing 1mo ago
I’m kinda surprised Apple doesn’t do something like this. I would imagine it’s a lot easier with phones with LiDAR
34.
by syntaxing 1mo ago
I think the fun takeaway from this is that GPT 5.4 is probably 45B active parameters and GPT 5.6 Sol is closer to 50B.
35.
by syntaxing 1mo ago
Thanks for the update! I wonder if a Blackwell GPU would be noticeably faster. Which vendor did you end up using? I want to get a gigabyte one but thats be OOS for months.
36.
by syntaxing 1mo ago
I think the hardest part is how to define “waste”. Are you “wasting” your time if you go to a school event for your kids during the weekday? Is it a waste of time walking your kids to the bus? Is it a waste of time reading to my kids before
37.
by syntaxing 1mo ago
What speed do you get on this setup? Im tempted to use the same GPU.
38.
by syntaxing 1mo ago
The issue with “AI” is how toxic it feels. Feels grimy like social media. Just like social media, I haven’t introduced any of it to my young kids nor do I plan to until they’re teens. At most I give them access to local AI through home assi
39.
by syntaxing 1mo ago
Would I be surprised there’s bench maxing happening? Yes. But some users also use Q4 quantized and complain how dumb local models are.
40.
by syntaxing 1mo ago
Is there a reason why so many of these agent harness are written in node.js?
41.
by syntaxing 2mo ago
Prefilling Kimi K3 with opus is a super interesting idea. That being said, I absolutely hate this website layout
42.
by syntaxing 2mo ago
Because distribution prices are spread throughout the users. This is the case for policies that allow the rich to be completely off grid which is not the case here in the US (most cap off around 1500W).
43.
by syntaxing 2mo ago
Simply not true. There’s a ton of induction and electric coil ranges with knobs. Just look at the Bosch line.
44.
by syntaxing 2mo ago
No way. I cook everyday and induction is vastly superior than gas. The only time fire is vastly superior is when I use a wok which I have a stove I use outdoors. Most induction stoves have temperature maintained too. So it can keep your oil
45.
by syntaxing 2mo ago
I’ve been using induction since I have had kids and I enjoy it so much more than fire. That being said, I rent so I can only use 120V ones and the options are so limited here. Most of the small 120V ones make an obnoxious hum.
46.
by syntaxing 2mo ago
I work in embedded space. Just because it’s in hardware doesn’t mean you can’t “protect” it. Most modern software (regardless if it’s hardware or not) can be cryptophically signed.
47.
by syntaxing 2mo ago
I don’t get why this is an issue? You can run Claude/OpenAI SOTA models through Amazon bedrock. These weights have to live somewhere to run on Bedrock.
48.
by syntaxing 2mo ago
SOTA American models are not. SOTA Chinese models are. From a physics aspect, closed source models cannot be too far from open source ones in terms of size. There’s only so much you can squeeze out a B100 style cluster even with fancy Dflas
49.
by syntaxing 2mo ago
Honestly, this is starting to make more and more sense. SOTA models are starting to converge to certain architecture and capabilities. I wouldn’t be surprised we end up with a base model ASIC + “fine tune” card where it’s a physical LoRA st
50.
by syntaxing 2mo ago
For personal stuff, I use it with AnythingLLM. It replaced any Google search for me. For coding, I run opencode though I have been debating switching to Pi. I would argue it’s at Sonnet 3 level.
51.
by syntaxing 2mo ago
I meant Qwen3.6. Unsloth supposedly has early preview of the model and the VRAM requirement is the same so most people expect similar model size and type.
52.
by syntaxing 2mo ago
I am so excited for Qwen 3.8 27B. It’s a shame how slow prefill (~3-400) is on a strix halo but it’s such a good model for agentic tasks.
53.
by syntaxing 2mo ago
This reminds me of Three body problem and how the scientist discovered the high strength wire was through quick physical experiments and use them as input to an AI model to determine if it works.
54.
by syntaxing 2mo ago
> View pricing in the Cloudflare dashboard ↗ Why… I wanted to see if it’s worth it to use cloudflare’s endpoint but I can’t even see the pricing
55.
by syntaxing 2mo ago
Interesting take on this https://aftermath.site/war-crime-n64-palmer-luckey-m64-modre...
56.
by syntaxing 2mo ago
I have a framework desktop, but depending on your need, DGX spark might be better. The prefill and NVFP4 is a significant advantage. But framework desktop is a better general computer. I expect to be able to use it for years to come. Where
57.
by syntaxing 2mo ago
I use it with a strix halo server. 35B runs stupidly fast. 27B is about 700 TPS prefill and 30 TPS token generation. Which interestedly is about what Kimi K3 gives me depending on provider.
58.
by syntaxing 2mo ago
I’m a mechanical design engineer (10 years) turned software engineer (7 years in autonomous vehicles). Is there a particular role you would recommend for someone with my experience?
59.
by syntaxing 2mo ago
As a renter myself, I get the author's pain. I ended up just installing this non destructive retrofit switch for most of my home automation + smart bulbs. [1] https://www.thirdreality.com/products/smart-switch-gen3
60.
by syntaxing 2mo ago
Same, I thought it was about some cool network tunneling that made the GPU look like native hardware on the Linux host.

More ›