spacy

15.7K posts

spacy banner
spacy

spacy

@dosco

building ax, dspy in typescript. researching systems of llms and compilers + llms. proud vibe coder

Katılım Temmuz 2008
1.6K Takip Edilen4.1K Takipçiler
spacy
spacy@dosco·
guy talking about GPUs in space when asked about connectivity causally mentions that starlink exists. space internet is now a give like home wifi, wonder what else we can put up there
English
0
0
0
85
spacy retweetledi
ₕₐₘₚₜₒₙ
ₕₐₘₚₜₒₙ@hamptonism·
YOUR LACK OF URGENCY IS WASTING YOUR POTENTIAL .
English
66
1.4K
10.8K
325.2K
Autism Capital 🧩
Autism Capital 🧩@AutismCapital·
Life pro tip: You can now disable Shorts in the YouTube app Go to settings (the top right gear icon) —> time management —> daily limits —> shorts feed limit —> select 0 minutes. Boom. Shorts gone. Hallelujah. Nature has healed. 🙏🏼
Autism Capital 🧩 tweet media
English
233
182
3.7K
255K
spacy
spacy@dosco·
@allTheYud @dylanmatt interesting how your belief also proves a second belief you must hold that scaling will get us there since we have nothing else in the realm of research and reality
English
0
0
0
42
Eliezer Yudkowsky
Eliezer Yudkowsky@allTheYud·
@dylanmatt Eg, Cotra in 2021 put out a 157-page report predicting AGI 30 years later. Even that late, but still before the ChatGPT Moment, that was the local socially optimal amount of edgy. Its error and reasons for overestimation were explicitly called in advance, but not within EA.
English
5
0
41
2.1K
spacy
spacy@dosco·
open source is a cache for code tokens validated by big human models that know their shit. and it's growing faster and more alive than ever
Bailey Pumfleet@pumfleet

Open source is dead. That’s not a statement we ever thought we’d make. @calcom was built on open source. It shaped our product, our community, and our growth. But the world has changed faster than our principles could keep up. AI has fundamentally altered the security landscape. What once required time, expertise, and intent can now be automated at scale. Code is no longer just read. It is scanned, mapped, and exploited. Near zero cost. In that world, transparency becomes exposure. Especially at scale. After a lot of deliberation, we’ve made the decision to close the core @calcom codebase. This is not a rejection of what open source gave us. It’s a response to what risks AI is making possible. We’re still supporting builders, releasing the core code under a new MIT-licensed open source project called cal. diy for hobbyists and tinkerers, but our priority now is simple: Protecting our customers and community at all costs. This may not be the most popular call. But we believe many companies will come to the same conclusion. My full explanation below ↓

English
0
0
6
643
spacy
spacy@dosco·
heard a pod on ai infra in space, it's cool now bit flipping from cosmic rays is not a big issue with stocastic systems like models.
English
0
0
0
78
spacy
spacy@dosco·
when people say models can't do something and assume it's a safe space, they're usual thinking one-shot, all models are about sampling and verification they can keep trying for longer, correcting towards the result
English
0
0
3
123
spacy
spacy@dosco·
you don't need big models for agents the new age small models are solid enough at code, combined with a agent substrate like ax-agent which only uses code and supports error correction they're cheaper and faster even with a high error rate.
English
0
1
1
105
spacy
spacy@dosco·
dpo will need trl and a bit more resources but it’s still sub 100 usd
English
0
0
0
131
spacy
spacy@dosco·
unsloth for sft needs a couple hours on 1 h100 for most useful tunes
English
1
0
1
149
spacy
spacy@dosco·
people are sleeping on easy it is to collect traces from a RLM agent like ax agent and tune a model like gemma 4 26b or qwen 2.5 9b. it’s all code gen traces easy to judge and etc none of that llm api function calling bs
English
3
3
33
1.4K
Tarun Sachdeva
Tarun Sachdeva@tarunsachdeva·
first time using axllm in production, really nicely done @dosco
English
1
0
3
155
spacy
spacy@dosco·
@trydotworks that’s what i mean by traces the sys prompt + chat context for each agent turn
English
1
0
1
87
erik@try.works
[email protected]@trydotworks·
@dosco Not traces but the entire input sent to the agent for each stage - it all happens through documentation, not prompts. The agent also generates the docs.
English
1
0
1
105
spacy
spacy@dosco·
@RuiHuang_art i'd project it on a wall in a loop and imagine we're already there
English
0
0
0
166
Rui Huang
Rui Huang@RuiHuang_art·
Will you watch 30 min of this?
English
63
102
1.2K
32.1K
spacy
spacy@dosco·
the rise of agents requires massive scaling of data and search infrastructure. every gpu will will control thousands of more cpus
English
0
0
5
181