lmeyerov

12 ms·

121.
by lmeyerov 9mo ago
You may enjoy Graphistry (eg, pygraphistry, GraphistryJS), where our users regularly do 1M+ graph elements interactively, such as for event & entity data. Webgl frontend, GPU server backend for layouts too intense for frontend. We have
122.
by lmeyerov 9mo ago
Data poisoning at its finest, wow
123.
by lmeyerov 9mo ago
I've had an interesting comparative experience so far using Alloy, which is a 'simpler' formal method approach that you can think of as smarter exhaustive fuzzing for your properties of interest. We have been building GFQL, a
124.
by lmeyerov 9mo ago
Thanks! Their discussion is super relevant to exactly what I wrote -- * They note speed benefits * The quality benefit they note is synonym search... which agentic text search can do: Agents can guess synonyms in the first shot for you, eg,
125.
by lmeyerov 9mo ago
Claude code / codex which internally uses ripgrep, and I'm unsure if it's using parallel mode. And, project specific static analyzers. Studies generally show when you do agentic retrieval w/ text search, that's pret
126.
by lmeyerov 9mo ago
We find across our team different people are able to use these things at different levels. Unsurprisingly, more senior coders with both more experience in general and more experience in ai coding are able to do more with ai and get more amb
127.
by lmeyerov 9mo ago
Was guessing that, coding tasks are a valuable but myopic lense :) I'm guessing a self-updating plan there is sufficient. I'm not actually convinced today's current plan <> todolist flow makes sense - in the linked PLAN
128.
by lmeyerov 9mo ago
I'm optimistic most people can, given the time and resources In the CCC video, you may enjoy the section on how we are moving to eval-driven AI coding for how we more methodically improve agents. Even more so, the slides before on moti
129.
by lmeyerov 9mo ago
Curious what kinds of evals you focus on? We're finding investigating to be same-but-different to coding. Probably the most close to ours that has a bigger evals community is AI SRE tasks. Agreed wrt all these things being contextual.
130.
by lmeyerov 9mo ago
Yep -- one fun experiment early in the video is showing sonnet 4.5 -> opus 4.5 gave a 20% lift We do a bit of model-per-task, like most calls are sending targeted & limited context fetches into faster higher-tier models (frontier but
131.
by lmeyerov 9mo ago
yep! So all repos get a (.gitignore'd) folder of `plans/<task>/plan.md` work histories . That ends up being quite helpful in practice: calculating billable hours of work, forking/auditing/retrying, easier repl
132.
by lmeyerov 9mo ago
I find in coding + investigating there's a lot of mileage to being fancier on the todo list. Eg, we make sure timestamps, branches, outcomes, etc are represented. It's impressive how far they get with so little! For coding, I act
133.
by lmeyerov 9mo ago
Something I would add is planning. A big "aha" for effective use of these tools is realizing they run on dynamic TODO lists. Ex: Plan mode is basically bootstrapping how that TODO list gets seeded and how todos ground themselves w
134.
by lmeyerov 9mo ago
Companies like xbow and horizon are using agents that talk to symbolic tools to automate more red teaming flows for different domains, so very much so. As shown in my talk, modern models are quite capable, and they aren't doing investi
135.
by lmeyerov 9mo ago
Thank you, and happy to answer questions on that, it's been a crazy time! Maybe of relevance to non-security people here: 1. Most of it is about AI investigating event data in general, not just SOC/IR: cyber, intel, fraud, SRE, an
136.
by lmeyerov 10mo ago
Most of this rings true for us for the same reasons. We have been moving large old projects in this direction, and new ones start there. It's easier to do these via tool checks than trust skills files. I wouldn't say the resulting
137.
by lmeyerov 10mo ago
Deleted, I'm not sure what's the point of talking with people about for/against Trump at this point
138.
by lmeyerov 10mo ago
I've been using Claude code most of the year, and codex since soon after it released: It's important to separate vibes coding from vibes engineering here. For production coding, I create fairly strict plans -- not details, but seq
139.
by lmeyerov 10mo ago
Now that I can do a lot with 3-6 AI agents running usefully 2-5min at a time to crank through my plans, the IDE is mostly just taking valuable space For backend/application code, I find it's instead about focusing on the planning
140.
by lmeyerov 10mo ago
Interesting, mind sharing the context here? My experience has been as workloads get heavier, it's "cheaper" to push to an accelerated & dedicated inferencing server. This doesn't always work though, eg, world of diff
141.
by lmeyerov 10mo ago
A few things Table stakes for our bigger users: - parity or improvement on perf, for both CPU & GPU mode - better support for learning (fit->transform) so we can embed billion+ scale data - expose inferred similarity edges so we can
142.
by lmeyerov 10mo ago
We generally run UMAP on regular semi-structured data like database query results. We automatically feature encode that for dates, bools, low-cardinality vals, etc. If there is text, and the right libs available, we may also use text embedd
143.
by lmeyerov 10mo ago
Fwiw, we are heavy UMAP users (pygraphistry), and find UMAP CPU fine for interactive use at up to 30K rows and GPU at 100K rows, then generally switch to a trained mode when > 100K rows. Our use case is often highly visual - see correlat
144.
by lmeyerov 11mo ago
I don't see news spread, eg, direct lineage graphs showing viral attribution & rewrites as a narrative propagates.. Afaict, it is the usual topic trending over time, or maybe it is showing direct sindication? Computing actual deriv
145.
by lmeyerov 11mo ago
I can believe, so a different question as the attribution is unclear: For context: A bunch of whitehat teams are using agents to automate both red + blue team cat-and-mouse flows, and quite well, for awhile now. The attack sounded like norm
146.
by lmeyerov 11mo ago
There is a big push to limit what kind of models can be OSS'd, which in turn means yes, a limit to what AI you are allowed to run. The California laws the article references make OSS AI model makers liable for whatever developers &
147.
by lmeyerov 11mo ago
(have been a big fan of this work for years now) From the nearby perspective of building GFQL, an embeddable oss GPU graph dataframe query language somewhere between cypher and duckdb/pandas/spark, at an even higher-level on top o
148.
by lmeyerov 1y ago
Is this embeddable, eg, a react component that can be hooked into? The lack of this has been a sticking point making us lean to dropping mermaid, so very cool to see!
149.
by lmeyerov 1y ago
Share can go up and down if consumption keeps going up crazily. We now spend more per dev on their personal use inferencing providers than their home devices, so inferencing chips are effectively their new personal computers...
150.
by lmeyerov 1y ago
Right, the 'knowing' is where I think the interesting thing is today for their evolution more mature claude.md files already typically index into other files, including guidance which to preload vs lazy load. However, in practice

More ›