serjester

11 ms·

61.
by serjester 1y ago
I love Ramirez and everything he’s done for Python, but this seems like another PaaS that isn’t really doing anything new. Does anyone other than a complete beginner (who probably isn’t paying) struggle to deploy a fastapi app in 2025?
62.
by serjester 1y ago
I appreciate you dropping those. In https://pubmed.ncbi.nlm.nih.gov/9855498/ the average dose was a staggering 440mg a month (multiple rolls a month) with a sample size of 24 people. This is definitely falls under &quo
63.
by serjester 1y ago
Ask anyone that's just starting out with geo pandas about their experience, and I'd be shocked if anyone calls it intuitive and straightforward. To geopandas credit, I think they just inherited many of Pandas' faults (why do
64.
by serjester 1y ago
I think this is part of a broader trend of geospatial data just becoming easier to work with. DuckDB is great for quick ad hoc stuff, but I find polars to be easier to maintain. Personally, I'm really excited for polars to (eventually)
65.
by serjester 1y ago
I would love to see some studies on this because everything I've seen is either rats that were exposed to truly insane doses (10X more than a human would take) or among long term, heavy users (weekly). I don't necessarily doubt th
66.
by serjester 1y ago
This seems like a step towards some sort of basic ad supported model. At the end of the day that's the only way they'll be able to survive long term against the money printing machine that is Google. Do I love the idea? No. But i
67.
by serjester 1y ago
I think this such an awesome innovation but does anyone understand what incentivizes a ship’s captain to take the warnings seriously? Is this just reliant on people being good people? I haven’t worked with sailors but I did a lot of blue co
68.
Gemini 2.5: The First LLM That Understands PDF Layouts(sergey.fyi)
16 points by serjester 1y ago|1 comments
69.
by serjester 1y ago
Anecdotally o3 is the first OpenAI model in a while that I have to double check if it's dropping important pieces of my code.
70.
Gemini 2.5 Might Close RAG's Transparency Gap(sergey.fyi)
2 points by serjester 1y ago|0 comments
71.
by serjester 1y ago
Just ran it on one of our internal PDF (3 pages, medium difficulty) to json benchmarks: gemini-flash-2.0: 60 ish% accuracy 6,250 pages per dollar gemini-2.5-flash-preview (no thinking): 80 ish% accuracy 1,700 pages per dollar gemini-2.5-
72.
by serjester 1y ago
Point of feedback on the landing page - you guys need to a better job explaining the killer use case here. Your first video is an implementation detail not a selling point. Your second video is asking how to establish a C Corp using contex
73.
by serjester 1y ago
Have you looked at fine tuning linear adaptors to sit on top of the embedding models? This works with any model (proprietary or open) and I think in practice this is significantly easier to implement anyways.
74.
by serjester 1y ago
I think anyone that cares enough about embedding performance to use niche models is probably parsing their PDF's into some sort of textual format. Otherwise you need orient your all your pipelines to handle images which adds significan
75.
by serjester 1y ago
Chunking is less important in the long context era with most people just pulling in top 20 K. You obviously don’t want to butcher it, but you’ve got a lot of room for error.
76.
by serjester 1y ago
Pivoting every few months? 0 cofounder alignment? Decisions like these are tough to make in the moment, but looking back you almost never regret it. Trust your gut.
77.
by serjester 1y ago
Bold to launch this before the roll out their own reasoning models / deep research. Seems like table stakes if you want to capture power users but maybe that's just my workflow.
78.
by serjester 2y ago
Are you worried about OpenAI and every other big lab eventually doing this? It’s going to be hard to get anyone to hand over this kind of data / control without a giant name attached.
79.
by serjester 2y ago
Even operator's original demo the first thing they showed was booking restaurant reservations and ordering groceries. I understand their need to demo something intuitive but it's still debatable whether these tasks are ones that m
80.
AI agents: Less capability, more reliability, please(sergey.fyi)
423 points by serjester 2y ago|253 comments
81.
by serjester 2y ago
Agreed when the metric becomes the goal, it stops being a useful metric. College attendance seems to fall in that bucket.
82.
by serjester 2y ago
Agreed. In college I would always go to the first class, see if the lecture was useful and probably 80% of the time I wouldn’t go again. Although I do sympathize with many of the author’s broader points.
83.
by serjester 2y ago
I think I have trouble understanding what this is doing other than maybe some fine-tuned prompts tailored to a specific stack? I'm looking at the data science kit and I don't see why anyone would use this, much less pay for it? I
84.
by serjester 2y ago
Agreed. Sundar seems like a peacetime CEO that got pulled into war and seems like he's struggling. But with Larry and Sergei having all voting control it seems unlikely he'll get fired.
85.
by serjester 2y ago
Has anyone met Googlers that are confident in the company's AI strategy? Anecdotally, everyone I've talked to seems to have serious concerns but that might just be a small sample size.
86.
by serjester 2y ago
Now someone needs to build semantic search over this to find hidden gem authors on any topic.
87.
by serjester 2y ago
I wish they’d mention pricing - it’s hard to seriously benchmark models when you have no idea what putting it in production would actually cost.
88.
AI Agents: Less Capability, More Reliability, Please(sergey.fyi)
3 points by serjester 2y ago|0 comments
89.
by serjester 2y ago
This misses that if the agent is occasionally going haywire, the user is leaving and never coming back. AI deployments are about managing expectations - you’re much better off with an agent that’s 80 +/- 10% successful than 90 +/-
90.
by serjester 2y ago
Synthetic data is just as useful for building app layers evals. Probably significantly cheaper ways to get the data if you’re training your own model.

More ›