jryio

7 ms·

61.
by jryio 7mo ago
Hey Oskar ~ great project and looks promising. I would be curious to hear what is still work-in-progress for Bombadil. It's helpful to know what the tool maintainers see as upcoming or incomplete work. It also saves a consultant like m
62.
You are not your job(jry.io)
382 points by jryio 7mo ago|408 comments
63.
by jryio 7mo ago
Not if having a heart attack within 1 year at a higher rate is an co-morbidity factor when the primary treatment was for obesity or diabetes (not stating that obesity and heart disease are not positively correlated). To use a dense analogy:
64.
by jryio 7mo ago
We finally found the first morbidity signal of GLP-1s (or lack thereof). These are life changing drugs, but like plastic we'll see their effects in force within this generation: > The longer time spent off GLP-1s, the greater the ri
65.
by jryio 7mo ago
This is the key piece > Full AttnRes is straightforward but requires O(Ld) memory at scale. Block AttnRes partitions layers into N blocks, accumulates within each block via standard residuals, and applies attention only over block-level
66.
by jryio 7mo ago
> The most positive outcome I can think of is one where computers get really good at doing, and humans get really good at thinking. If we never figure out how to make computers creative, then there will be a very natural division of lab
67.
by jryio 7mo ago
Businesses exercise power and control in the market. The purpose of this is to set a precedent (perceived or actual) — the auth system was a product, not an API. Anthropic is drawing the line between 'built on us' and 'built
68.
by jryio 7mo ago
There are not 8 major versions between iOS 18 and iOS 26. Apple skipped the monotonously increasing version numbering system since iOS 1 during WDDC 2025 to adopt a year suffix based versioning system. iOS 17, then iOS 18, then iOS 26, then
69.
by jryio 7mo ago
Here is the Google Research group's writeup https://cloud.google.com/blog/topics/threat-intelligence/dar... Relevant forward: > GTIG has identified several different users of the DarkSword exploit cha
70.
Assign tasks to Claude Cowork from anywhere(support.claude.com)
3 points by jryio 7mo ago|0 comments
71.
by jryio 7mo ago
Historians will write about the 'metaverse' as the last software frontier humanity embarked on before AI. What a shame. Hopefully capitalism and AI research does not produce equally bad products and ideas.
72.
by jryio 7mo ago
Correct. Where did the engineering go? First it was in code files. Then it went to prompts, followed by context, and then agent harnesses. I think the engineering has gone into architecture and testing now. We are simply shuffling cognitive
73.
by jryio 7mo ago
This is a naïve approach, not just because it uses FizzBuzz, but because it ignores the fundamental complexity of software as a system of abstractions. Testing often involves understanding these abstractions and testing for/against the
74.
Language model teams as distributed systems(arxiv.org)
104 points by jryio 7mo ago|46 comments
75.
by jryio 7mo ago
Correct
76.
by jryio 7mo ago
You must explicitly state what your threat model is when writing about security tooling, isolation, and sandboxing. This threat model is concerned with running arbitrary code generated by or fetched by an AI agent on host machines which con
77.
by jryio 7mo ago
1 million tokens is great until you notice the long context scores fall off a cliff past 256K and the rest is basically vibes and auto compacting.
78.
by jryio 7mo ago
This appears to be published by Google itself
79.
by jryio 7mo ago
The matte. It's offensive.
80.
by jryio 7mo ago
I own this as well and while I appreciably the levelized cost, there is simply zero comparison to my gen 1 Studio Display. The gloss and shin on the Kuycon means it only works in dimmly lit rooms. Nano texture in mixed lighting scenarios is
81.
by jryio 7mo ago
¥¥¥
82.
by jryio 7mo ago
All code has bugs, the vector space of all possible bugs determines the entropy of the problem space for a large language model to traverse. Reduce entropy, increase probability of the correct outcome. LLMs are surfing higher dimensional ve
83.
by jryio 7mo ago
I've been saying this for maybe nine months vis-à-vis my consulting work keeps proving it. Go is an excellent language for LLM code generation. There exists a large stable training corpus, one way to write it, one build system, one for
84.
by jryio 7mo ago
This an appropriate rewind to unreasonable behavior. I applaud Anthropic's candor in the public sphere. Unfortunately the country party is unworthy of such applause.
85.
by jryio 7mo ago
Without circular investments and valuations what would Open AI be worth? 100B? 300B? Entirely on revenue alone it seems like 20B. Current valuation appears to be two orders of magnitude off.
86.
by jryio 8mo ago
This is interesting for offloading "tiered" workloads / priority queue with coding agents. If 60% of the work is "edit this file with this content", or "refactor according to this abstraction" then low lat
87.
by jryio 8mo ago
There have been a few of these kanban[1] user interfaces over claude code or some other agent (open claw). These are all proto orchestrators. No one has discovered/converged on what agent orchestration actually looks like. Other projec
88.
by jryio 8mo ago
These comments are comical. How hard is it to understand that human beings are experiential creatures. Our experiences matter, to survival, to culture, and identity. I mourn the horse masters and stable boys of a century past because of the
89.
by jryio 8mo ago
Accountants will be studying the deals and cyclical valuations of AI companies in the same way we study bank runs and FDIC insurance today.
90.
by jryio 9mo ago
Think about how it feels when you toil on a hard problem, do your best work, release it to the work in the spirit of openness and sharing Only to have a machine ingest, compress, and reiterate your work indefinitely without attribution.

More ›