efromvt

7 ms·

61.
by efromvt 4mo ago
A self para-sql-lang plug if you like the direction of malloy but prefer something closer to native SQL syntax. (a controversial take at times) - https://trilogydata.dev/
62.
by efromvt 4mo ago
I think advanced SQL authoring is generally simple to understand, and that's the larger learning curve! I find those big stored procedures usually fall into two categories; logic that should be in the DB, but should be decomposed (stag
63.
by efromvt 4mo ago
Interesting - which models specifically? I'd be interested in using mistral over deepseek if it was competitive (guess I need to go benchmark)
64.
by efromvt 5mo ago
Incredible how reliable the heuristic of "something seems off - probably github being down" has gotten these days
65.
by efromvt 5mo ago
For my streamlined, composable SQL variant - Trilogy[1] - spent the last month or two very focused on CLI/ETL/scheduling to support data pipelines for other hobby projects. Have that story in a better place - so time for new thing
66.
by efromvt 5mo ago
I don't the OpenClaw furor has been a problem for the majority; but stuff like the harness bugs with dropped thinking traces (capacity optimization?) and some fairly bizarre billing bugs with weird/opaque comms around both have be
67.
by efromvt 5mo ago
that is a terrible way to figure out if openclaw is used, hah
68.
by efromvt 5mo ago
Can you still build a garage in the bottom corner to get more starting money??
69.
by efromvt 5mo ago
I didn’t realize they had chDB, honestly, need to give that a shot! The local CLI isn’t quite the same ergonomically
70.
by efromvt 5mo ago
Sqllite and duckdb serve pretty different niches; duckdb is less embeddable but on the OLAP side it’s by far the best today. I wouldn’t ever see them as competing for the same app, though
71.
by efromvt 5mo ago
I try to avoid jumping on the bandwagon when it's already covered but billing bugs being treated like other software issue and the major comms channel being X (which I can't get to load half the time) is ridiculous.
72.
by efromvt 6mo ago
Love the layering approach - that solves a problem I’ve had with other sql/visual hybrids as you move past the basics charts.
73.
by efromvt 6mo ago
Hmm yeah it depends on your definition of insider. If you assume all raw information is public-ish, a market can reward those who can synthesize/operate on that knowledge to predict better. (The cars in the lot, etc. there is friction
74.
by efromvt 6mo ago
Still working on my urban tree visualization ! Spent some time polishing the ingest pipeline to make it easier to add new cities, added a genus/species level view to aggregate across cities, and added in some basic imagery so I can see
75.
by efromvt 6mo ago
I wonder why this doesn’t get us frustrated with the grid, not data centers. Delays on interconnects for renewables and offshore wind both seem pretty self inflicted.
76.
by efromvt 6mo ago
I’ve been curious about this too - obvious performance overhead to have a internal/external channel but might make training away this class of problems easier
77.
by efromvt 6mo ago
Yeah the marginal cost of discovery going towards 0 (I mean, not there yet, but directionally) is the problem; it doesn't really matter if the agent isn't equivalent to a human artistic hand-crafted bug discovery if it can make it
78.
by efromvt 6mo ago
Should I read that as 'generic system'? Most hard data is with company internal evals, but for the well defined tasks externally it's been pretty easy to spin up a basic tool loop and validate. Did you have something in mind?
79.
by efromvt 6mo ago
Just like OpenAI's original moat, I don't think that's particularly durable. I've already seen plenty of people swing back to preferring codex, and it'll probably swap again with the next model drop. Openclaw is pot
80.
by efromvt 6mo ago
It's pretty easy to get determinism with a simple harness for a well-defined set of tasks with the recent models that are post-trained for tool use. CC probably gets some bloat because it tries to do a LOT more; and some bloat because
81.
by efromvt 6mo ago
I haven't heard this benefit for mentors clearly articulated before (probably just missed it), but definitely felt it - I guess it's a deeper version of how writing/other communication forces clarity/organization of thou
82.
by efromvt 6mo ago
I think this is exactly the point though (maybe more of the link than of this comment) - a sufficiently good product by all external quality metrics is fine even if the code is written on one line in a giant file or some other monstrosity.
83.
by efromvt 6mo ago
I've chose to embrace the silver lining where there is now business backing to prioritize all the devx/documentation work because it's easier to quantify the "value" because LLM sessions provide a much larger sample
84.
by efromvt 7mo ago
The gemini models are fantastic for price but the naming scheme is ridiculous, I have to triple check it every time.
85.
by efromvt 7mo ago
Inference costs at least seem like the thing that is easiest to bring down, and there's plenty of demand to drive innovation. There's a lot less uncertainty here than with architectural/capability scaling. To your point, tomo
86.
by efromvt 7mo ago
Yeah I have a 'species' info table that's built by curating wikipedia and a few other sources and passing them through a structured LLM pipeline; ecological benefit; blooming season; native regions, etc. This is very much a &
87.
by efromvt 7mo ago
I actually 100% agree with your for a new DBMS and share your frustration with vendor-specific features and lock-in. At that level, it's often actively counterproductive for insurgent DBs - ecosystem tooling needs more work to interfac
88.
by efromvt 7mo ago
As long as you don't deviate too much from ANSI, I think the 'light sql DSL' approach has a lot of pros when you control the UX. (so UIs, in particular, are fantastic for this approach - what they seem to be targeting with qu
89.
by efromvt 7mo ago
This is where I think we need better tooling around tiered validation - there's probably quite a bit you can run locally if we had the right separation; splitting the cheap validation from the expensive has compounding benefits for L
90.
by efromvt 7mo ago
If you found two disjoint sections that seemed positive on their own, did you try looping both separately in the same model? Wondering how localized the structures are.

More ›