@someRandomDev5

Shipping to prod in a $100M+ ARR unicorn startup w/ local AI. 4x DGX Spark (GLM 5.3 Flash), 1x M5 Max 128GB (Qwen 3.8 Flash Next), 1x RTX 5090 (Qwen 3.8 27B).

Joined February 2018
  • Tweets32.5K
  • Following80
  • Followers541
2.9KMedia
Include

Only show posts containing:

Exclude

Hide posts containing:

Time range
-
Minimum likes
Replying to @MiaAI_lab @exploit_intel
Shows significantly higher prefill.
5
Replying to @levelsio
I don't even know what this is. We don't use it. Is this something that vibe coders use?
90
Random Libertarian Tech Lead retweeted
成龍
@graphtraversal8h
1
47
4
614
8,936
> 69.3% of the web runs on PHP And only 30.7% of the web gets more than 1 visit per month.
@levelsio
@levelsio19h
I pre-emptively disconnected ALL my sites / VPS servers from the web I'll wait for the patch to hit at 2 PM UTC If you have a PHP website, probably good to take them down until you patch PHP 69.3% of the web runs on PHP so this is going to be intense! Good time maybe to buy cybersecurity stocks? (not financial advice)
8
Replying to @h1kz0r @ciprianveg
I don't know that I would buy at the $7k price point. But at the $3k-$4k price point, my 4x setup was great, and I could've seen myself upgrading to 8x in the near future.
5
Replying to @someRandomDev5 @h1kz0r @ciprianveg
And you can run w/e abliterated / fine-tuned GLM 5.3 model you want on your own hardware at this rate. You can red-team your own apps w/o worrying about whether the ability to do so will be regulated away. Remember when Anthropic suddenly pulled Fable after its first 3 days?
2
31
Replying to @someRandomDev5 @h1kz0r @ciprianveg
GLM 5.3 has roughly the same tokens/outcome efficiency as Opus 5, so the tokens/sec rating pretty much translates 1:1 to "actual outcome speed".
1
2
26
Replying to @h1kz0r @ciprianveg
Well: 1. EXL3 4bpw isn't "retarded". 2. For now, the hardware prices continue to climb, so it's the opposite of "negative ROI" if you sell sometime before they fall. 3. In terms of TPS, the 50-60tok/s range has been pretty standard until recently. Example:
2
19
Today's PHP vulnerabilities are the outcome of decades of human slop.
1
1
29
Replying to @h1kz0r @ciprianveg
As low as $24k if you bought Asus GX10s for their launch price of $3k each late last year. Even as recently as August, GX10s were $4k each, putting you at $32k for 8 for them. So "what did I miss" = All of the decent prices before the dramatic price swing lol.
1
27
Replying to @alexhvnsen
> Every experienced vibe coder I know, knows how to make the output look more natural. Lmao, no they don't. They just think they do.
6
Replying to @Yanxbolts @jeebus27 @BreidenFehoko
He's a Super Bowl winner and an NFL MVP.
1
13
Replying to @TGatliff @MiaAI_lab @Recursive_SI
Lol. I used it more after this, and let me tell you: It's pretty indicitive of its intelligence relative to other models.
8
Replying to @TGatliff @MiaAI_lab @Recursive_SI
Yeah, it's a tool. But in this case, it's a tool that's not intelligent enough to abide by the system prompt and guidance that tells it what an @-tagged file means. It got lost despite simple instructions that were spelled out explicitly.
1
11
Replying to @someRandomDev5 @TGatliff @MiaAI_lab @Recursive_SI
To be clear, the former one would eventually find it too... Once it finishes searching literally the entire volume for a tagged file lol. 😂
1
10
Replying to @someRandomDev5 @TGatliff @MiaAI_lab @Recursive_SI
But if you tell it "uh it's definitely right here and it exists" then it can find it lol.
1
1
9
Replying to @someRandomDev5 @TGatliff @MiaAI_lab @Recursive_SI
Straight up tagged in Pi and it can't find it lol.
1
1
5
Replying to @someRandomDev5 @TGatliff @MiaAI_lab @Recursive_SI
And the first time I tried running it with their suggested harness, the model did a full long `find` command to try to locate a file that was literally tagged with the full path lmao.
2
1
5
Replying to @TGatliff @MiaAI_lab @Recursive_SI
Not even close. Doesn't even make it up to 27B's level. xitter.cf/cline/status/211296712…
Cline
@cline14h
We’re about to see a breakout in local model capabilities. Flint scores 54.9% on Terminal-Bench 2.1 with ~2B active parameters. That’s ahead of every other small MoE and two of the three dense ~30B models. It only needs about 24GB and runs at 125+ tok/s on a MacBook Pro.
1
1
39
Replying to @someRandomDev5 @kentcdodds
Use the AI for what it's exceptional at. Use yourself for what the AI sucks at. stuwe.dev/blog/knowing-think…

White-Collar Work: Knowing, Thinking, Doing

White-collar work is knowing, thinking, and doing. LLMs excel at two of these. Here's how to structure your workflow to maximize both human and AI contributions.

stuwe.dev
2
Load more