Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
hyluo
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
4 ms
·
1.
▲
Show HN: Subconscious and GLM-5.2 Makes "/compact" Obsolete
(subconscious.dev)
2 points
by
hyluo
3mo ago
|
0 comments
2.
▲
Subconscious Cache for Agent Inference
(subconscious.dev)
2 points
by
hyluo
4mo ago
|
0 comments
3.
▲
by
hyluo
6mo ago
Looks pretty cool! I love the automatic trigger functionality
4.
▲
Agent Engine Optimization (AEO): Selling to AI Agents
(github.com)
1 points
by
hyluo
7mo ago
|
0 comments
5.
▲
Show HN: Single-agent long-horizon reasoning within one LLM run
(huggingface.co)
4 points
by
hyluo
1y ago
|
1 comments
6.
▲
End-to-end long-horizon reasoning with one Transformer model
(subconscious.dev)
5 points
by
hyluo
1y ago
|
3 comments
7.
▲
by
hyluo
1y ago
- We build the Thread Inference Model (TIM) based on the transformer architecture, and its dedicated runtime TIMRUN. - TIM + TIMRUN = Intelligent workflow generation, context engineering, and multi-hop tool use happens at the runtime level
8.
▲
SmoothGPT on GPTs to avoid ChatGPT detecters
(chat.openai.com)
1 points
by
hyluo
3y ago
|
1 comments
9.
▲
by
hyluo
3y ago
thoughts?
10.
▲
by
hyluo
3y ago
The paper introduces improved performance by prompting LLMs with "natural language embedded programs (NLEP)". No task-specific prompt is needed. Paper: https://arxiv.org/abs/2309.10814 An automatic NLEP gener
11.
▲
Program generation is all you need? For math, symbolic, natural language, etc.
(arxiv.org)
1 points
by
hyluo
3y ago
|
1 comments
12.
▲
by
hyluo
3y ago
Thanks! will replace "chatgpt" with "gpt-3.5-turbo".
13.
▲
by
hyluo
3y ago
It is a problem, but the same training set can be used to train commercial models.
14.
▲
by
hyluo
3y ago
The model outputs ChatGPT by connecting to a search engine - it's expected because it is fed with search results. Without search, SAIL-7B does not outperform ChatGPT. The interesting aspect is that the ChatGPT response quality is not i
15.
▲
by
hyluo
3y ago
Sry our backend gpu servers are being upgraded and it caused some problem for our job submission system. Our slurm is not fixed yet, but I've brought the model back online for now - it will be temporally diabled again when our slurm is