lmeyerov

33 ms·

451.
by lmeyerov 3y ago
Agreed, I think the monopoly scenarios play out similarly/worse to the lock in ones -- monopolies are basically the extreme case of lock in. They're the ultimate removal of end-user influence on purchase decisions. Frustratingly,
452.
by lmeyerov 3y ago
The article's stance is those are effects, not causes: As a product succeeds and establishes lockin, management may decide that devoting resources to improving the product's design has less relative ROI than devoting those resourc
453.
by lmeyerov 3y ago
This practice being considered acceptable has been a nightmare - we had OSS libraries add spying later, change their APIs for disabling it, etc. When it's a nested dependency, even worse. Now and then we run our stack with mitm monitor
454.
by lmeyerov 3y ago
So not OSS?
455.
by lmeyerov 3y ago
groq has been public on launching an api, so running groq+mixtral vs openai.com (vs others) is a real topic, esp. when ~half of openai's revenue is their api
456.
by lmeyerov 3y ago
Very much this. It's incredible what we can do with it - our limiter building louie.ai has been more ourselves than OpenAI. Though of course a better LLM (smarter, faster, cheaper) does enable more. The main problem with freezing is th
457.
by lmeyerov 3y ago
Earth-1, where we are happily already sharing parquet and arrow files for years now that do that :) The subtle bit is most languages & these formats have data representation (int16, ..) but not actual unit (GB/s) and oh boy....
458.
by lmeyerov 3y ago
If you find this stuff useful, I recommend also checking out our pygraphistry umap() for a few reasons here: - by visually linking similar entities, vs just a scatter plot of values floating around or binning, you see a lot of the substruct
459.
by lmeyerov 3y ago
GPUDirect is amazing, I wish more of our (Graphistry's) customers wanted it. We've had surprisingly few deploys like that yet one of the only ways we can do 100+ GB/s per-node analytics Are you using GDS or networking or both
460.
by lmeyerov 3y ago
... and we did, added to our Jaeger/Prometheus. Works great!
461.
by lmeyerov 3y ago
Agreed w others, a workaround is see if any other VCs want your pro rata at some discount, where they front you the money and you agree on some split. The startup likely has right of first refusal, which begins a negotiation. Presumably if
462.
by lmeyerov 3y ago
Re:python, if we are already doing otel, how would this interop? Eg, if we don't want to break our current imports, and control where the new instrumentation goes (Fwiw, This is a great direction!)
463.
by lmeyerov 3y ago
All these seem historically bizarre, and making me have deep suspicion about the people & processes driving this legislation. I expected California to legislate early, but not like this. 1. What I expected: * AI Indemnification rules fo
464.
by lmeyerov 3y ago
Until the founder gets tired, replacement CEO gets hired, and with VC board encouragement, sells to PE or BigCo who switches the defaults 18mo later. Users get the choice of dead, CVE-riddled software or following the company's structu
465.
by lmeyerov 3y ago
+ this It was super painful transition for Graphistry. Certainly as a founder responsible for payroll and having only so many hours in a day. Likewise, as business where people in serious gov, enterprise, etc teams are making mission bets o
466.
by lmeyerov 3y ago
+1 We do a variant of gpu umap+hdbscan => viz + scores a LOT. We are big fans! We've grown in 2 pretty different directions to berttopic to support our users investigations over the last few years: -- At least for our users, before
467.
by lmeyerov 3y ago
Graphistry | Viz engineer | Security engineer | Contract | REMOTE - GLOBAL | https://www.graphistry.com + https://www.louie.ai Graphistry is a Gartner-awarded startup used by fraud/cyber/BI/supplychain
468.
by lmeyerov 3y ago
Setting aside the nonsense missions, I find it surprisingly useful from a self-determination perspective when working with a strongly motivated team: - On one hand we say '100Xing investigations'. This is customer centric, focusin
469.
by lmeyerov 3y ago
Ouch, first Disney/Pixar, now Microsoft / Activision Blizzard For #webgl folks ready to beyond entertainment & ads, we may be able to help turn lemons into lemonade: - Opening: Senior webgl infoviz engineer to help us build th
470.
by lmeyerov 3y ago
Sponsor, even!
471.
by lmeyerov 3y ago
Related: If any OSM GIS engineering enthusiasts, we have a small OSS repo we are looking for an assist on that should be fun for the right person: Spin up a big box, quickly render planet-scale tiles at a decent level (maptiler), and restar
472.
by lmeyerov 3y ago
Tesseract OCR fallback sounds great! There are now a lot of file loaders for RAG (langchain, LLMindex, unstructured, ...), any reasons, like a leading benchmark score, to prefer this one?
473.
by lmeyerov 3y ago
We experimented with this: Typically we get nice small contributions from devrel teams around connectors in their software, and junior data scientists newer to coding who we encourage to do examples instead of core code We revisit this ever
474.
by lmeyerov 3y ago
It's funny taking a numbers view. The most popularly used might not even be these, but vector indexes in existing popular OSS DBs and storage systems people are already using. Afaict earliest would be faiss on disk and vectors in opens
475.
by lmeyerov 3y ago
I agree with those perspectives, they're questions we talk about and analyze: - We do in fact operate multiple OSS repos. Despite their usage, most, as with most OSS in general, do not get significant code-based contributions. Our prop
476.
by lmeyerov 3y ago
Fair question! A good chunk would cancel, and at a minimum, switching to free sw / paid support would drop our product-related revenue by ~70-80%. Then again, we could make the software worse to trigger higher support hours, as complai
477.
by lmeyerov 3y ago
There is some tricky big assumption being made here around sustainable profitability that misses our lived reality, especially given challenges like US developer salaries. Paraphrasing, OSS companies need lightning to hit twice, first for t
478.
by lmeyerov 3y ago
`lgtm` is a legit typical outcome. biasing to approving, in many companies & software setups, is a great default. ex: we might `lgtm`, or make comments that are suggestions for follow-on PRs without blocking the current. However... it&#
479.
by lmeyerov 3y ago
Hah that was me ~12 years ago trying to get WebCL (OpenCL) through the same gate keepers. Meanwhile, in Python, we are doing multi-node multi-GPU. Maybe OpenAI's and soon Apple's success with LLMs will change the economics for the
480.
by lmeyerov 3y ago
MADD on one end that limit work at all, and SOC2/ISO/NIST style compliance limitations on the other that chill work in practice. I'm more worried about the latter as already starting to bias project & funding decisions in

More ›