m_ke

10 ms·

1.
by m_ke 21d ago
You have to understand that Gates had other distractions to deal with
2.
by m_ke 21d ago
just like there were mistrals, coheres, llamas, etc, there will be new deepseeks and moonshots if those ever flame out (worst case, given out at cost by google, meta, alibaba or etc) OpenAI and Anthropic are already in a ~200bil hole from p
3.
by m_ke 21d ago
So how exactly is Anthropic and OpenAI ever going to pay back the trillions that they plan on spending?
4.
DuckDB Labs is joining Amazon [video](youtube.com)
6 points by m_ke 22d ago|1 comments
5.
by m_ke 1mo ago
With spec decode should easily get to >100tps on my dual 3090s qwen 3.5 27b was running at around 110tps using the config from https://github.com/noonghunna/club-3090 make that 200tps on a single 5090, 4x faster tha
6.
by m_ke 1mo ago
when you're waiting for network or LLM inference the raw performance doesn't matter at all
7.
by m_ke 1mo ago
1. it's built for async 2. runs everywhere 3. interpreted, making it fast to iterate on 4. decent performance 5. most popular language, llms are decent at writing it
8.
by m_ke 2mo ago
All of these harnesses should support pinning config files and tools to specific models or families of models. It's really tiring to have to tweak everything with each model release and then watch those changes mess up cheaper models i
9.
by m_ke 2mo ago
GLM 5.2 feels better than Opus and K3 is as good as Fable. Now I can't wait for someone to distill K3 into a Qwen 3.6 27b or Poolside S 2.1 sized models for a proper fast local Composer 2.5 replacement.
10.
by m_ke 2mo ago
Yeah some real main character energy from Dario as usual. I'll never get why he thinks China would just sit there and let the US dominate them in AI when all it would take is a few of their boats blockading Taiwan to put a stop to it a
11.
by m_ke 2mo ago
the argument is that we should all fold and let sam altman burn trillions of dollars on naive scaling and pay monopoly prices for their closed APIs until the models are good enough to be closed off for "safety" reasons so that the
12.
by m_ke 2mo ago
Oh this definitely happens all the time. I was an early employee at Clarifai, which won imagenet a year after alexnet and we were able to stay on the frontier for about 2 years before a bunch of open source models were matching our results.
13.
by m_ke 2mo ago
only if you only get your news from main stream business press and Big Lab propaganda channels There's no chance K3 is a distill of Fable, it came out way too soon after the limited fable release to be feasbile. If you look at all of t
14.
by m_ke 2mo ago
no, the goal was to spark a conversation about the value of *open AI* and it looks like it worked
15.
by m_ke 2mo ago
Also open sourced a bunch of infra to go with it. Anyone who claims open source and open weights models are "decel" needs to get their head checked https://github.com/MoonshotAI/MoonEP https://gith
16.
by m_ke 2mo ago
On Policy Self Distillation and Active Learning. Anything that increases sample efficiency by providing a richer more dense feedback signal and is more efficient at exploration / sampling.
17.
by m_ke 2mo ago
I wonder what they'll do when there are open european and american models that lap them. Will we see "Open"AI claim open ai is a threat to humanity?
18.
by m_ke 2mo ago
Anthropic should think hard about all their fear mongering. It will only end up backfiring on them and everyone else involved. They definitely used closed private saas products to train their own models, to prove that just drop random small
19.
by m_ke 2mo ago
I assumed google would lean into the efficiency stuff more and try to eat the easy 80% of workloads, winning market share on volume instead of frontier if they were not able to produce frontier level models. They're very well equipped
20.
by m_ke 2mo ago
k3 costs will go down at least 3x within a week of the weights dropping. we'll get new quants, dspark speculators, distills and optimized kernels as long as there are near frontier models available there will be inference providers sel
21.
by m_ke 2mo ago
Sure, but anthropic is charging businesses based on usage now and tried hard to pull Fable from the consumer subscriptions before Sol and K3 dropped. Even now on the $200 plan I use up my Fable credits in a single day and had to start using
22.
by m_ke 2mo ago
Anthropic will get squeezed by open models for 80% of the use cases that don't require frontier capabilities and by vertical specific labs for the high value tasks that would (bio, finance, math, etc.), where smaller use case specific
23.
by m_ke 2mo ago
most of the gains from the past year and a half have not been from web data, but from synthetic data and agent rollouts with RL.
24.
by m_ke 2mo ago
yeah I tried rules, hooks and forbidding things like pip / python3 but it just led to the model failing to do what it wants and a bunch of token churn trying to get around my more rigid constraints. main problem is that the harness fil
25.
by m_ke 2mo ago
It's not an issue of usual vs unusual, I'm saying the models are way better at writing and maintaining django or react code bases than your own hand rolled architecture that you define in some docs that it has to learn and keep in
26.
by m_ke 2mo ago
I tried really hard to do this, but it turns out the models don't care about your ideas and want to do what's popular in their training data, so they will happily ignore anything you try to force down their throats, especially as
27.
by m_ke 2mo ago
With agentic coding the costs of tokens compound with each message / tool call and etc. Having to load in and update large files makes things slower and way more expensive. Databricks actually just posted some of their own benchmarks o
28.
by m_ke 2mo ago
After running out of Fable credits in a day on my max plan I started looking around for ways to trim down my token usage and came to the realization that all of the type spaghetti that opus wrote is probably eating up like 50-70% of my toke
29.
by m_ke 2mo ago
Deep Learning models are designed to get max throughput on GPUs, which ends up being batched workloads. You'll never get proper price competitive utilization on personal hardware vs a cloud inference provider that can batch and pipelin
30.
by m_ke 2mo ago
Model is the product, people will put up with anything as long as you're on the pareto frontier of performance x cost.

More ›