lmeyerov

27 ms·

391.
by lmeyerov 2y ago
We are long-time fans of Caddy, preferring it over traefik + nginx especially for our docker-compose flows.. though it's fair to distinguish 'can' vs 'easy to do' E.g., we can imagine writing or using a plugin to fi
392.
by lmeyerov 2y ago
For Caddy, LetsEncrypt: Free TLS in one line without talking to anyone For Traefik, afaict, something about k8s
393.
by lmeyerov 2y ago
Yes. You can think of Reasoning and Knowledge seperately, and sense their limits on these, to be able to guess for yourself: 1. Big LLMs are good at reasoning. Given a word problem, they can turn it into arithmetic. So they can generate pse
394.
by lmeyerov 2y ago
A good PhD isn't just about learning how to solve problems but also how to quickly ramp up & identify worthwhile ones. To give a feel, I might have read 20-200 papers for every new one that I came up with, I'd write a few a ye
395.
by lmeyerov 2y ago
I spent some time there. The Uber and Block offices are beautiful, as is Twitter's up the street. Unfortunately, the street is meth hell even before night hits. The city granted nice discounts for the CEOs to take the spots when rents
396.
by lmeyerov 2y ago
Having co-founder DNA in the market is easiest, and second best is being open minded (ex: young and naive, a PhD, ...) and immerse yourself in the customer space. We did the latter, and it took us ~4 years to match our inhouse deep tech wit
397.
by lmeyerov 2y ago
Graphistry / Louie.AI | MULTIPLE: Principal Security Researcher / Senior Security Sales Engineer | US+REMOTE | FULL-TIME + CONTRACT | https://www.graphistry.com/careers Graphistry is a Gartner-awarded visual graph
398.
by lmeyerov 2y ago
We've been working on gov/enterprise/etc projects here as part of louie.ai that are coming to a head -- I'd expect not far from what XAI, Perplexity, and others are also doing. However, while those must focus on staying
399.
by lmeyerov 2y ago
Yes! My position on KGs largely flipped post GPT-3. Before KGs were mostly a niche thing given the cost vs rewards, and now they're an everyone thing. - The effort needed for KGs has higher general potential value because RAG is useful
400.
by lmeyerov 2y ago
AFAICT most of the "graph" rag implementations discussions, instead of fancy graph queries & or structured knowledge graph, mean: 1. Primary: Inverted index on keywords (= entities). At ingest time, extract entities and revers
401.
by lmeyerov 2y ago
I don't follow. Maybe the point is a lot of people are not balancing their systems, so by sticking with wimpy-era IO architectures, they're not feeding their GPUs? I think about balancing nodes differently when designing older Spa
402.
by lmeyerov 2y ago
Combining BERT + RGCNs is great. Transformers on text + categorical features, then GNNs for learning over connected data, esp their choice of RGCNs for heterogeneous ones. Some of my favorite use cases to do here are in entity resolution &#
403.
by lmeyerov 2y ago
Sort of Databricks etc architectures are mostly slowly moving data from A to B and doing little work, and worse when that describes the distributed compute part too. I read a paper awhile back where that was often half the time. GPU archite
404.
by lmeyerov 3y ago
In a recent interview, Mark Zuckerberg said they're spending $10B-$100B on training and inferencing over the next X years. They see open source as a way to get the community to cut that cost. In his view, even just 10% cheaper inferenc
405.
by lmeyerov 3y ago
Yeah, we built it to essentially stream columnar record batches from server GPUs to browser GPUs with minimal touching of any of the array buffers. It was very happy-path for that kind of fast bulk columnar processing, and we donated it to
406.
by lmeyerov 3y ago
An irony here is that we donated the pure JS/TS implementation of arrow to apache
407.
by lmeyerov 3y ago
Last we checked for one of our use cases around sandboxing, key pydata libraries were slowly moving there, but it takes a village. At that time, I think our blockers were Apache Arrow, Parquet readers, and startup time. There were active is
408.
by lmeyerov 3y ago
I like to think of Open Governance vs Open Source on topics like, if someone adds SSO to the OSS repo, will the main vendor behind it reject the PR because it breaks their OSS core business model
409.
by lmeyerov 3y ago
Extra fun: We find most enterprise/gov graph analytics work only requires 1-2 attributes to go along with the graph index, and those attributes often are already numeric (time, $, ...) or can be dictionary-encoded as discussed here (ca
410.
by lmeyerov 3y ago
As someone who worked in the field, I think there were 3 big directions in the last ~20 years that remain relevant: * Experiments in interface design, the equivalent of language design. These balance being able to guide the system and being
411.
by lmeyerov 3y ago
Sort of... the IDE would want the mypy (or otherwise) typings to surface that. Internally, the dataframe library should make it easier for the IDE to see that, vs today's norm of tracking just "Any" / "Index" &
412.
by lmeyerov 3y ago
This is neat, I like the direction! As far as I can tell, it's runtime, not static, so it won't help during our mypy static checks period? As intuited by the poster above, we already do generally stick to Apache Arrow column types
413.
by lmeyerov 3y ago
Unless I'm misunderstanding, Arrow solves the data representation on disk/memory, both for pandas and polars, while I'm writing about type inferencing during static analysis, which Arrow doesn't solve. Having a type chec
414.
by lmeyerov 3y ago
Where I really want this is pandas. The community has been smoothing the basic typing story over the last couple of years, which helps with deprecations & basic API misuses. However, I'm excited for shape/dependent typing over
415.
by lmeyerov 3y ago
Encryption at rest of local terraform state seems beyond the threat model & compliance needs of a pre-PMF consumer startup It's a lot harder to be an enterprise company or b2b startup (SOC2) and brush that off in front of knowledge
416.
by lmeyerov 3y ago
Interesting! The 'much incentive' part is the interesting thing to me. Hashicorp did build encryption-at-rest for state into their paid remote cloud service. They explicitly recommend using the paid service instead when security i
417.
by lmeyerov 3y ago
Do you think a feature like this is tied to distancing OpenTofu from Hashicorp, e.g., AWS & friends have less incentive to do enterprise feature gating around security?
418.
by lmeyerov 3y ago
Very cool! Well-written, and surprisingly real: Somewhere VERY close to this modeling patterns has always been an interesting use case for us is large-scale visual analysis of crypto investigations, where the ledger gets shown quite similar
419.
by lmeyerov 3y ago
Yep, we do precisely this, st + SVM => pre filtering / routing => RAG pipelines for some of louie.ai In our cases, < 100 labels on modern models (per task) goes far. We can do better, but more useful problems to solve. Impres
420.
by lmeyerov 3y ago
It's a fascinating project, and esp for from before docker+k8s normalized and nodejs was reinventing the universe Nowadays, docker compose or k8s ecosystems do a lot of this stuff in a more team-friendly & ecosystem-supported manne

More ›