lmeyerov

18 ms·

241.
by lmeyerov 2y ago
My favorite here was we found Google's assigned ads consultant, when we dug into their glowing numbers, got us a big traffic spike in school kids using google's search bar as a graphing calculator. We are a b2b graph analytics com
242.
by lmeyerov 2y ago
To a reasonable rounding error.. yes
243.
by lmeyerov 2y ago
Blockchains are a fantastic way to run things slowly ;-) More seriously: Making crypto fast does sound like a fun technical challenge, but well beyond what our finance/gov/cyber/ai etc customers want us to do. For reference,
244.
by lmeyerov 2y ago
Just sanity checking here - with flight write streams to duckdb, I'm guessing there is no notion of transactional boundary here, so if we want data consistency during reads, that's another level of manual app responsibilities? And
245.
by lmeyerov 2y ago
Exactly this, they are so close!
246.
by lmeyerov 2y ago
'it varies' b/c we do everything from an interactive analytics chat agent (loiue.ai UI) to data-intensive continuous-monitoring (louie.ai pipelines) to one-off customer assists like $B court cases 1. Common themes in our deve
247.
by lmeyerov 2y ago
Agreed, and that's where teams like the OP come in OpenAI does great at training for general tasks, and we should not be disappointed when specialized tasks fail. Interestingly, openai advertises increasingly many subjects they are sp
248.
by lmeyerov 2y ago
I'm more confused now. If this is a tough and high-value task, we would not use gpt-4o-mini on its own, eg, add more steps like a verifier & retry, or just do gpt-4o to begin with, and would more seriously consider fine-tuning in
249.
by lmeyerov 2y ago
I like the direction, but have a pretty different experience in practice. This spans legal analytics, social media analytics, code synthesis, news analysis, cyber security LLMs, etc: 1. The only ultimate absolute quality metric I saw in tha
250.
by lmeyerov 2y ago
Yeah sorry -- I mean paid dues in diff ways, such as study and practice (earned), and having seen the realities of the business / software / last attempts (lived).
251.
by lmeyerov 2y ago
Yep very contextual as replies show! Ex: I'd not expect a code commit to happen at a bank or most gov places, eg, you might be waiting for clearance for months to even see the real data! There are many ways to set the tone, and code d
252.
by lmeyerov 2y ago
Assuming not management track: - Make a token commit day 1 or 2 to show you can and will deliver, even if just docs. Ensure you can go end-to-end on a real code PR week 1, even if a trivial one. Identify which area of code is most important
253.
by lmeyerov 2y ago
We evaluated the pyodide ecosystem for louie.ai last year and too much didn't work, eg, we couldn't get data in for arrow or parquet, which are table stakes for python data work My guess was 2025 would be more practical for these
254.
by lmeyerov 2y ago
Graphistry, Louie. AI | Director of R&D; GPU AI Research Fellow | https://www.graphistry.com https://www.louie.ai | US Remote Graphistry powers data-intensive investigations for enterprises, governments, NGOs, an
255.
by lmeyerov 2y ago
May help to think of these tools as on the opposite end of the spectrum. As an analogy: 1. langchain, llamaindex, etc are the equivalent of jquery or ORMs for calling third-party LLMs. They're thin adapter layers with a bit of consiste
256.
by lmeyerov 2y ago
Are you doing some regular like vision? For the reasons you're saying, I don't think it's an accident that GNNs are popular mostly in domains like recommendations that feel graph-y for their domain model so getting to a usefu
257.
by lmeyerov 2y ago
It may help to think about baselines. The status quo is being late or bad, understaffed, low budget, etc. It is not comparing to perfect code written by top CS grads with all the time in the world. Developers need help writing basic code, a
258.
by lmeyerov 2y ago
At the federal level, most things are big, underfunded, understaffed, underpaid, and over regulated, so programs are low quality, behind, and slow. I stay away from state level, and expect there is less data and $, and even more problems. T
259.
by lmeyerov 2y ago
The presumption of the article is that software correctness is the same the formal methods (which I lump mainstream classical program synthesis correctness into) is the same as some symbolic software satisfying some strong logical predicate
260.
by lmeyerov 2y ago
I don't feel like you're reading what I'm writing. I regularly use LLMs to work with security protocols and encrypted data, which are wrapped in APIs, and write new GPU code. Likewise, I regularly use LLMs to more quickly wri
261.
by lmeyerov 2y ago
I would welcome DOGE focusing on how much velocity it will increase and how many lines of bad policy it will remove. Likewise, cutting inefficient people should instead be replacing them. It is so hard to make progress within gov, so focu
262.
by lmeyerov 2y ago
Her keynote at last year's https://www.louie.ai/gen-ai-graph-the-planet/ explicitly ended on this Basically the way forward she's most excited about wrt AI x Security as a way avoid devolving into AIs fightin
263.
by lmeyerov 2y ago
See my above on there is an academic use of terminology here that makes any discussion here irrelevant. A big chunk of the article reduces to pedantic trolling by basing arguments on this.
264.
by lmeyerov 2y ago
Not sure I understand What % of programmers in your company writes verifying coq proofs or equiv for what goes to production? Vs what % just does testing (unit, integration, etc)? For us, as a small team, we do 7-8 figure annually for teste
265.
by lmeyerov 2y ago
I'm mostly responding to the middle third of the article that IMO interprets the word "correct" in an academic way that, IMO, is unanchored to how, to a rounding error, software is written: > "Even simple versions of
266.
by lmeyerov 2y ago
It's frustrating whenever folks throw up proof complexity as why LLMs can't work. If most programs most people want to write can map into predictable & verifiable abstractions, or we recognize almost no software is verified to
267.
by lmeyerov 2y ago
How is the community doing authentication & authorization? We're considering how to support in louie.ai, anticipating a large ecosystem of sites supporting it, yet the original spec had essentially TODO for how to do authentication
268.
by lmeyerov 2y ago
Agreed! The bigger shift is switching to data parallel coding styles.
269.
by lmeyerov 2y ago
Agreed!
270.
by lmeyerov 2y ago
Comparing multicore wide AVX to CUDA is a bit of an unnecessary nuance for most folks. These make sense, but miss the forest from the trees: - Either way, you're writing 'cuda style' fine-grained data parallel code that looks

More ›