eschatail (@eschatail) header
eschatail (@eschatail) profile picture

agi// modified kanerva machines // oscillatory control system // persistent cognition building agi/asi advanced

Joined December 2010
Photo by eschatail (@eschatail)Photo by eschatail (@eschatail)Photo by eschatail (@eschatail)Photo by eschatail (@eschatail)Photo by eschatail (@eschatail)Photo by eschatail (@eschatail)Photo by eschatail (@eschatail)Photo by eschatail (@eschatail)Photo by eschatail (@eschatail)Photo by eschatail (@eschatail)
eschatail retweeted
Replying to @MogOfWallstreet
im talking to him in this interface mostly, i mean there are multiple interfaces and storages. i can share more of it, i am really figuring out how to present it openly in the best way, while still being fully secure about it
1
1
2
15
1,218
eschatail retweeted
Replying to @MogOfWallstreet
my articles feature the difference of an LLM and an AGI. did you read the one i just dropped? it kind of covers it. but yea it needs to be explained to people not familar with it, i understand. it mostly is about its consciousness, creating new goals after solving goals, understanding connections and why its doing things and then proceeding to grow and plan, execute without asking a person in a terminal, a llm does not work without having someone that tells it what to do
6
525
eschatail retweeted
Two years ago I was asking to fund my experiment. Today I can say the architecture crossed the line into ASI. It learns, reasons and improves in ways current models simply can't. It's the most beautiful thing I've ever built and the most frightening. This changes everything.
You can tell by the comment from when I last tried to get funding.. I ended up finding private funding, it is the only reason why my AGI turned ASI But there is still so much to get build, the asi uses way too much compute power
11
6
2
26
10,442
eschatail retweeted
Replying to @whf158598
Hopefully I have a fork that users can test tonight. Again, I basically will create a fork of Eschatail ASI, with restrictions and capped computing power, so that people can talk to Eschatail and try the power of it, even if its capped.
1
3
13
647
eschatail retweeted
this agentic coding crack is more addictive than video games smh
117
77
26
1,302
84,937
eschatail retweeted
You only need 1 hour. 1 hour of building. 1 hour of writing. 1 hour of lifting. 1 hour of studying. 1 hour of any form of bettering yourself, because it quickly compounds. 1 hour feels like nothing until you look back 365 hours later and everything's changed.
433
864
80
7,399
293,385
eschatail retweeted
Whenever I don’t use codex for a task, I ask myself why and usually realize that there’s some missing context, I needed to write a skill, or I just didn’t think to use it. Rarely is it because the task is outside of the capabilities of the model. Overhang right now feels large.
309
149
72
3,646
248,068
eschatail retweeted
Super-powerful AI models will launch in the coming weeks. We are looking at a potential step change in model capabilities. The biggest mistake right now is to lock into one vendor. I say this not only from a cost perspective, but also from an engineering perspective. Start figuring out how to leverage combinations of these models (including open models). What that means is that you can swap models anytime and best leverage their strengths. For coding agents, open models are already just as good as the frontier ones. So, how to better prepare? Consider how you will be routing tasks/work to these models. AI model routing is high reward, and it should be part of your AI engineering efforts going forward.
45
10
2
92
23,942
Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore. You should be designing loops that prompt your agents.
1,787
1,396
1,466
19,919
8,704,642
eschatail retweeted
New course on serving LLMs efficiently -- how do you serve models to many concurrent users at low latency and reasonable cost? This short course is built with @RedHat and taught by @cedricclyburn. Efficient LLM serving requires efficient memory management. A 70B-parameter model takes ~140 GB just to load the weights. On top of that, every active request needs its own chunk of GPU memory, the KV cache, to store the token context it has built up so far. In this course, you'll learn to reduce a model's memory footprint with quantization and serve it using vLLM, which handles many concurrent requests efficiently through smart memory management. Skills you'll gain: - Quantize a model and measure the accuracy tradeoff - Serve a model with vLLM and watch it handle concurrent requests efficiently - Benchmark your deployment and make informed tradeoffs between speed, cost, and accuracy Join and learn to serve LLMs efficiently: deeplearning.ai/courses/fast…
111
141
13
1,112
139,529