lmeyerov

17 ms·

181.
by lmeyerov 1y ago
I came from a town where farmers work with university scientists and the ones who don't are basically doing tourist farm stands. Likewise, I see the NSF + NIH funding cuts from fellow scientists, eg, cancer researchers, being cut, and
182.
by lmeyerov 1y ago
You might enjoy Lotus: https://github.com/lotus-data/lotus It takes all the core relational operators and makes an easy semantic version of each as a python dataframe library extension . Each call ends up being a &#x27
183.
by lmeyerov 1y ago
We're talking about different things. I agree CVE ratings and risk/severity/etc levels in general for third party libraries are awkward. I don't have a solution there. That does not mean we should stop reporting and trac
184.
by lmeyerov 1y ago
I agree on the first part, but it's useful to be more formal on the latter -- 1. Agreed it's totally fine for a system to have some bugs or CVEs, and likewise fine for OSS maintainers to not feel compelled to address them. If some
185.
by lmeyerov 1y ago
Traditional security follows the CIA triad: Confidentiality (ex: data leaks), Integrity (ex: data deletion), and Availability (ex: site down). Something like SOC2 compliance typically has you define where you are on these, for example Does
186.
by lmeyerov 1y ago
I think bar for b2c, prosumer, SMB, yes.. folks want to see fast revenue growth (vs eyeballs & ideas)... but enterprise... not as much, and that's half the b2b market Reverse there afaict, enterprise + defense tech are booming. AI
187.
by lmeyerov 1y ago
That assumes they were implemented with deterministic operators, which isn't the default assumption when using neural network libs on GPUs. Imagine random seeds, cublas optimizations - like you can configure all these things, but I wou
188.
by lmeyerov 1y ago
For louie.ai, our users are doing agents and agentic reasoning for automating daily investigation work: 1. Agentic Automation: For every alert/ticket coming in, the agent does a pre-investigation across relevant APIs, DBs, etc, helping
189.
by lmeyerov 1y ago
I don't recall there being favorable reception when Trump's Congress passed it nor since. At best, non-awareness, and 'cost of business' for whatever the other talking points at the time. You can see the older HN threads
190.
by lmeyerov 1y ago
If/when to commit prompts has been fascinating as we have been doing similarly to build Louie.ai. I now have several categories with different handling: - Human reviewed: Code guidelines and prompt templates are essentially dev tool in
191.
by lmeyerov 1y ago
There is an interesting subtlety to what you are saying - * Inferencing is getting cheaper: 40-80% / yr in hardware + 40-80% / yr in software. Annual numbers will change, but for next 5 years, seems fine to assume continued imp
192.
by lmeyerov 1y ago
I agree they are compute limited and disagree that they are aggressively optimizing. Many small teams are consistently showing many optimization gain opportunities all the way from app to software to hardware, and deepseek was basically jus
193.
by lmeyerov 1y ago
Maybe this is the disconnect for the token side: you seem to think they can't keep improving the margin to reach profitability. They are static and it will just get worse, not better. I think deepseek instead just showed they haven&#x2
194.
by lmeyerov 1y ago
> But why? In the world you're positing where these queries are sold at negative margins and don't provide any other tangible benefit (i.e. cannot be used for training), the provider would be even better off not serving those q
195.
by lmeyerov 1y ago
I think you ignored both of my points - 1. You just negated a technical statement with... I don't even know what. Engineering opportunities at volume and high skill allow changing the margin in ways low volume and low capitalization pr
196.
by lmeyerov 1y ago
Interesting. I have been doing a simple man's version of multiple git clone folders and 'docker compose -p'. Making that smoother is attractive, esp if can be made opaque for our more junior teammates. On one end, I have bee
197.
by lmeyerov 1y ago
I think you miss 2 big aspects: 1. High volume providers get efficiencies that low volume do not. It comes from both more workload giving more optimization opportunities, and staffing to do better engineering to begin with. The result is br
198.
by lmeyerov 1y ago
I recently wrote a 5+ page internal guide on how I do vibe coding and one of the first sections is cost Keep in mind much of the guide is about how to move from 30s chats to doing concurrent 20min+ runs ---- Spending Claude Code $$$ - Max P
199.
by lmeyerov 1y ago
I like to think about maximizing throughput while minimizing attention: both matter, and the proposal here is expensive on my attention. Optimizing per-task latency matters less than enabling longer non-interactive runs. For parallelism, I&
200.
by lmeyerov 1y ago
Slightly different experience here We have been adding MCP remote server to louie.ai, think a semantic layer over DBs for automating investigations, analytics, and viz over operational systems. MCP is nice so people can now use from Slack,
201.
by lmeyerov 1y ago
Splunk Boss of the SOC is the realistic test, it is one of the best cyber ranges. Think effectively 30+ hours of tricky querying across 100+ real log source types (tables) with a variety of recorded cyber incidents - OS logs, AWS logs, aler
202.
by lmeyerov 1y ago
I can bring data here: We recently did the first speed run where Louie.ai beat teams of professional cybersecurity analysts in an open competition, Splunk's annual Boss of the SOC. Think writing queries, wrangling Python, and scanning
203.
by lmeyerov 1y ago
I started a company that makes a bunch of many every year based off of NSF research, and Nvidia is making even more based on our work that spun out. I'm not a unique story, yet they cut half the NSF budget and half the NSF-funded STEM
204.
by lmeyerov 1y ago
Yes, by default things should be sandboxed - no network, no repo writes, ... - and should be easy to add extra caps (ex: safelist dockerhub) Likewise, similar to modern smart phones asking if they should remove excess unused privs granted t
205.
by lmeyerov 1y ago
Louie.ai - security genAI | full-time, DC area | https://www.louie.ai Louie.ai is pioneering generative AI for data intensive investigations. This includes Vibes Investigating with AI notebooks, autonomous agentic investigations
206.
by lmeyerov 1y ago
We had similar earlier on at Graphistry. It was pretty obvious, especially by the time of video screens. We are still unsure if whether a hacker or just someone avoiding their history/nationality - online history was sparse and somewha
207.
by lmeyerov 2y ago
Yeah exactly We still want chunking in practice to avoid LLM confusion, undifferentiated embeddings, and handling large datasets at lower cost + large volumes. Large context means we can now tolerate multi-paragraph/page, so more like
208.
by lmeyerov 2y ago
I meant Roe as an example where the absence of federal law triggered a hybrid interpretive mode. Judges follow the plain meaning of the law, but when the text is missing, ambiguous, or leads to absurd results, they shift to hybrid methods b
209.
by lmeyerov 2y ago
Enterprise math is interesting -- For a global all-you-can-eat enterprise-wide rollout: * base: 20K users x $200/yr * 50% discount: volume + multi-year + ... => enterprise: $1M/yr => 200 person division in the enterprise: $1
210.
by lmeyerov 2y ago
I don't think you're disagreeing with me. The US system is letter-of-the-law, we agree there. The issue is when there is ambiguity. Judges must then make their own interpretation, and things get murky. In the US, spirit might be u

More ›