Hacker News: brilee

What's in a Codebase?

brilee — Sun, 05 Apr 2026 19:23:31 +0000

Article URL: https://www.moderndescartes.com/essays/codebase_spec/

Comments URL: https://news.ycombinator.com/item?id=47652925

Points: 2

# Comments: 0

Eight years of wanting, three months of building with AI

brilee — Sun, 05 Apr 2026 12:43:47 +0000

Article URL: https://lalitm.com/post/building-syntaqlite-ai/

Comments URL: https://news.ycombinator.com/item?id=47648828

Points: 582

# Comments: 179

What's in a Codebase?

brilee — Thu, 26 Mar 2026 15:54:57 +0000

Article URL: https://www.moderndescartes.com/essays/codebase_spec/

Comments URL: https://news.ycombinator.com/item?id=47532060

Points: 1

# Comments: 0

Interviewing for ML/AI Engineers

brilee — Fri, 26 Dec 2025 16:27:38 +0000

Article URL: https://www.moderndescartes.com/essays/ml_eng_interviewing/

Comments URL: https://news.ycombinator.com/item?id=46393532

Points: 1

# Comments: 0

New comment by brilee in "We Induced Smells With Ultrasound"

brilee — Sun, 23 Nov 2025 12:03:05 +0000

I could have sworn that sickly sweet smell was the smell of various phosphine reagents? Just my vague recollection of my time in lab from 15 years ago.

New comment by brilee in "Parsing Chemistry"

brilee — Wed, 05 Nov 2025 16:06:32 +0000

Does this handle, e.g., water of hydration CaSO4 . 2H2O? states of matter H2O(g)? does it preserve subunit information, as in (C6H5)CH2COOH? Writing a parser for basic formulae is such a tiny tiny part of the actual problem... deciding the scope of what you want to handle and how is the real problem

Taste

brilee — Wed, 29 Oct 2025 13:52:24 +0000

Article URL: https://www.moderndescartes.com/essays/taste/

Comments URL: https://news.ycombinator.com/item?id=45746841

Points: 2

# Comments: 0

New comment by brilee in "A bug that taught me more about PyTorch than years of using it"

brilee — Sun, 26 Oct 2025 12:50:51 +0000

Great write-up, but I admit that I found the interweaving of human and AI-written content/headlines/summaries pretty distracting. I kept on wanting to scroll past, but had to keep on backtracking to find the human thread again.

I think if you want to give your reader a quick intro to, e.g., what is the Adam optimizer, a simple link to Wikipedia is fine. No need to copy-paste an AI tutorial on Adam into the blog post.

Codebase is 250% AI generated

brilee — Fri, 17 Oct 2025 11:34:04 +0000

Article URL: https://www.moderndescartes.com/essays/ai_codebase/

Comments URL: https://news.ycombinator.com/item?id=45615494

Points: 16

# Comments: 12

New comment by brilee in "Protecting your (my) passport from liquid damage"

brilee — Mon, 06 Oct 2025 12:03:16 +0000

AI slop

New comment by brilee in "You Had No Taste Before AI"

brilee — Thu, 18 Sep 2025 14:35:34 +0000

Taste is the field behind the goalposts, and we as humans are constantly expanding that field by "having taste" and developing taste as a community.

https://www.moderndescartes.com/essays/taste/

Taste

brilee — Tue, 16 Sep 2025 18:40:54 +0000

Article URL: https://www.moderndescartes.com/essays/taste/

Comments URL: https://news.ycombinator.com/item?id=45266134

Points: 1

# Comments: 0

Taste

brilee — Fri, 12 Sep 2025 16:30:31 +0000

Article URL: https://www.moderndescartes.com/essays/taste/

Comments URL: https://news.ycombinator.com/item?id=45223872

Points: 2

# Comments: 0

New comment by brilee in "Deploying DeepSeek on 96 H100 GPUs"

brilee — Fri, 29 Aug 2025 16:05:34 +0000

For those commenting on cost per token:

This throughput assumes 100% utilizations. A bunch of things raise the cost at scale:

- There are no on-demand GPUs at this scale. You have to rent them for multi-year contracts. So you have to lock in some number of GPUs for your maximum throughput (or some sufficiently high percentile), not your average throughput. Your peak throughput at west coast business hours is probably 2-3x higher than the throughput at tail hours (east coast morning, west coast evenings)

- GPUs are often regionally locked due to data processing issues + latency issues. Thus, it's difficult to utilize these GPUs overnight because Asia doesn't want their data sent to the US and the US doesn't want their data sent to Asia.

These two factors mean that GPU utilization comes in at 10-20%. Now, if you're a massive company that spends a lot of money on training new models, you could conceivably slot in RL inference or model training to happen in these off-peak hours, maximizing utilization.

But for those companies purely specializing in inference, I would _not_ assume that these 90% margins are real. I would guess that even when it seems "10x cheaper", you're only seeing margins of 50%.

Multimodal LLMs Are Blind

brilee — Mon, 25 Aug 2025 14:35:41 +0000

Article URL: https://www.moderndescartes.com/essays/blind_llms/

Comments URL: https://news.ycombinator.com/item?id=45014322

Points: 3

# Comments: 0

Claude as Pipeline Orchestrator

brilee — Mon, 25 Aug 2025 14:35:16 +0000

Article URL: https://www.moderndescartes.com/essays/claude_pipeline/

Comments URL: https://news.ycombinator.com/item?id=45014316

Points: 1

# Comments: 0

New comment by brilee in "Deep-Sea Desalination Pulls Fresh Water from the Depths"

brilee — Sat, 16 Aug 2025 11:23:47 +0000

Desalinated water is also less dense than normal seawater, so the water column inside the output pipe would create a pressure imbalance with the water column outside the pipe, assisting in the outflow? I'm having trouble figuring out how to resolve this seeming perpetual motion machine

New comment by brilee in "Study mode"

brilee — Tue, 29 Jul 2025 20:58:43 +0000

I'm working on a startup in this space and wrote up my thoughts here: https://www.moderndescartes.com/essays/study_mode/

What ChatGPT Study Mode gets right and wrong

brilee — Tue, 29 Jul 2025 20:57:52 +0000

Article URL: https://www.moderndescartes.com/essays/study_mode/

Comments URL: https://news.ycombinator.com/item?id=44728190

Points: 3

# Comments: 0

New comment by brilee in "Anthropic Is Bleeding Out"

brilee — Fri, 11 Jul 2025 18:33:06 +0000

As a former big tech engineer, I can't help but come up with a gazillion ways to work around these sorts of seemingly straightforward policies.

Here's one way they could get around their own privacy policy: keep track of what % of Claude-generated code is retained in the codebase over time (as an indicator of how high-quality / bug-free the code was); A/B test variations of Claude Code to see which variations have higher retention percentages.

No usage data is retained, no code is retained, no data is used (other than a single floating point number) and yet they get to improve their product atop your usage patterns.

Here's another idea: use a summarization model to transform your session transcript into a set of bits saying "user was satisfied/dissatisfied with this conversation", "user indicated that claude was doing something dangerous", "user indicated that claude was doing something overly complicated / too simple", "user interrupted claude", "user indicated claude should remember something in CLAUDE.md", etc. etc. and then train on these auxiliary signals, without ever seeing the original code or usage data.