JacobAsmuth

4 ms·

31.
by JacobAsmuth 2mo ago
Isn't the top end of the distribution by definition what it's capable of?
32.
by JacobAsmuth 2mo ago
I would prefer that Trump keep all the power personally. He's done such great work recently I don't see why he shouldn't be in control of the machine god
33.
by JacobAsmuth 2mo ago
You're hallucinating facts again.
34.
by JacobAsmuth 2mo ago
I've said this for a long time. This inability of AIs to generate novel explanatory hypotheses is a big blocker for their ability to automate jobs like accountant, middle manager, and even the lowly cashier.
35.
by JacobAsmuth 2mo ago
The irony of using Claude to write this is almost to much to "hold". (Claude loves to tell you to hold things in its writing). Here's another great Claudism: "The replies had a different read. One of them, in full: {embe
36.
by JacobAsmuth 2mo ago
Really? You prefer not to see the top end of the distribution? Why?
37.
by JacobAsmuth 2mo ago
> mobile era CEO He was the guy branding Google an AI-first company back when they invented the transformer.
38.
by JacobAsmuth 2mo ago
So then it is similarly telling that compute is second on the list? They have more talent than compute? Layoffs incoming?
39.
by JacobAsmuth 2mo ago
Dallonians
40.
by JacobAsmuth 2mo ago
only 30 upvotes in 3 hours but 50 comments(many negative) is not a great signal.
41.
by JacobAsmuth 2mo ago
Me after I spend a day at the beach: #BA3A17
42.
by JacobAsmuth 2mo ago
Why do you think you've prevented third party supply train risk when you're training on a 30+ Terabyte blob that you have little understanding of its contents. It's essentially entirely composed of unfiltered user input.
43.
by JacobAsmuth 2mo ago
Can you give an example of a question that you would google, but your colleagues would "Ask Claude" and they get a hallucination but you get the right answer?
44.
by JacobAsmuth 2mo ago
1e21 flops is hilariously wrong. for reference the llama 3 8B model ( https://arxiv.org/pdf/2407.21783 ) used 10 times that many flops. This model is 350x bigger in total params and 12x bigger in active params and was tr
45.
by JacobAsmuth 2mo ago
Now you're thinking like management!
46.
by JacobAsmuth 2mo ago
Why do you feel that creating the LLM from scratch is sufficient to trust it? Are you suggesting that you personally would read all 15 trillion tokens (plus every single agentic trade used in RL, along with its relative advantage in the bat
47.
by JacobAsmuth 2mo ago
It's 3/15 - https://openrouter.ai/moonshotai/kimi-k3 If you're going to open source your model, why would you set your own price high enough that other providers could easily and profitably undercut you?
48.
by JacobAsmuth 2mo ago
Their competitors aren't able to serve models as good as Gemini 3.6 Flash to billions of users
49.
by JacobAsmuth 2mo ago
buddy man
50.
by JacobAsmuth 2mo ago
You're confused by this? Fable > Opus > Sonnet > Haiku. Sort by price. Most expensive = better. Are you an alien? lol
51.
by JacobAsmuth 2mo ago
All providers are equally trustworthy :)
52.
by JacobAsmuth 2mo ago
AA isn't the best way to measure relative cost in real world use because some of those benchmark questions are extremely hard for the models. Some models give up quickly on hard questions, other models spin their wheels for a long time
53.
by JacobAsmuth 2mo ago
What? Claude plays Pokemon one-shots the entire game without any harness other than claude code and game screenshots
54.
by JacobAsmuth 3mo ago
Marco econ 101 disagrees :) Ask your favorite LLM to explain the "yield curve" to you
55.
by JacobAsmuth 3mo ago
How has your experience been with Gemma 4?
56.
by JacobAsmuth 3mo ago
You need to compare cost per task buddy boy. Cost per token doesn't tell you much when you don't know how many tokens a model will use to accomplish a task
57.
by JacobAsmuth 3mo ago
Could it be that they have to serve their models to billions of users?
58.
by JacobAsmuth 3mo ago
Just you. This is typical market competition in a fast moving field.
59.
by JacobAsmuth 3mo ago
(Rate limits * capability of the model) / cost
60.
by JacobAsmuth 3mo ago
It's also about 15x faster.

More ›