adsharma

5 ms·

1.
by adsharma 4d ago
The problem is that you're not offered a choice. No one is making the "err towards explainability" choice. Training data is treated as IP. Distillation is seen as an attack. Open data, open training based systems such an Mari
2.
by adsharma 4d ago
What you're using is a transformer based database index. An obfuscated one. Training data is the database. Model is a lossy compressed index.
3.
by adsharma 4d ago
I'm sympathetic to this argument. All I'm saying is: if you have a choice between two systems with equal power of discovery and one is more understandable than the other, we choose the more understandable one. Limits of Human comp
4.
by adsharma 4d ago
Oh you want tech that helps discover new science instead of parroting existing wisdom? There is little evidence that the RSI we are discussing is capable of inventing the theory of relativity (or the more advanced equivalent). All we have s
5.
by adsharma 4d ago
If there is guillotine in the weights of the model, it needs to be properly labeled so you can look it up by name using a database index (or a graph-vector index). It helps both the bad guys and good guys. Like responsible disclosure in cyb
6.
by adsharma 4d ago
You can have a transformer based index built on top of a graph database. Much of the explainability tech (people who prefer "alignment" also prefer "mechanistic interpretability") is reverse engineering the real world gr
7.
by adsharma 4d ago
Statistical learned indexes can exist in explainable tech such as a database. In 2017 Google was writing papers about it. Then something changed. I don't think it was the tech. It was a realization around the power and societal impact.
8.
by adsharma 4d ago
The gun comes with a manual on how to use it safely. I'm sure it has some complexities, but at the high level: > A gun is a metal tube that uses a tiny, controlled explosion to shoot a small piece of metal (called a bullet) forward
9.
by adsharma 4d ago
The real threat is that we uncritically adopt language such as alignment. Implicit in this is the idea that AI is a inscrutable matrix and going to remain that way and we'll need expert interpreters to make sense of it. We need to insi
10.
by adsharma 4d ago
So many words. But missing the one that matters the most: Explainability. AI slowdown is worth it only if it can be made more explainable. Changing the language we use to discuss it is a good first step. We need to stop using inside basebal
11.
PostgreSQL to pull property graphs from v19 after design flaws(freenode.net)
5 points by adsharma 7d ago|0 comments
12.
by adsharma 14d ago
I'm encouraged by this result. It's the primary hypothesis behind latentpedia.org. Instead of distilling the geometry of a model into a huge knowledge graph, we start from the largest known open source graphs and build it up towar
13.
by adsharma 16d ago
Simplicity is in the eye of a beholder. It's not an argument backed by facts. Many single file graph databases out there. They should be "simpler" since it's one file vs many. Text files vs binary is what you're arg
14.
by adsharma 20d ago
Many competent embedded graph DBs out there. More suitable than graphs on SQLite projects.
15.
by adsharma 20d ago
For what benefit though? Much better off storing this in an embedded graph database and use cypher. MVCC, Indexes, Strong typing, WAL, changefeeds. The simplicity of markdown and JSON are deceptive. SQLite is a solid choice, but then you de
16.
by adsharma 21d ago
https://www.youtube.com/@ladybugdb - what content would you like to see?
17.
by adsharma 21d ago
Several "Graph on DuckDB" efforts started on DuckDB and ended up reinventing a basic columnar codebase to innovate on because of these reasons. Even though people didn't document why, lack of outside contributor friendly flow
18.
by adsharma 21d ago
Surprised that there isn't more discussion of DuckDB source distributions. Why do we need a source distribution for a well regarded MIT licensed project? Because it's not easy to contribute code to DuckDB if you don't work at
19.
by adsharma 22d ago
I measured this on a M4 mac mini (base model): zig build sqlite-benchmark Medium (100K nodes) +----------------------+-----------+---------+---------+ | Workload | LatticeDB | SQLite | Speedup | +-----------------
20.
by adsharma 22d ago
Wikidata doesn't imply RDF/SPARQL. Cypher works fine too. A columnar storage engine means you get indexes and the relational goodness for free. https://huggingface.co/datasets/ladybugdb/wikidata-20260401
21.
by adsharma 22d ago
LadybugDB maintainer here. Couple of corrections: * LadybugDB has revamped the Kuzu WAL design. It shouldn't be hard to build WAL based replication * 19ms vs 39us - like the author says these are vastly different systems and the benchm
22.
by adsharma 22d ago
https://huggingface.co/datasets/ladybugdb/github-knowledge-c... These methods could not beat the 480 token result because of the small dataset. But for a sufficiently large corpus with graph reordering and columna
23.
by adsharma 22d ago
Did you consider extracting knowledge from the doc into a graph database (while passing the Q&A) and compressing the graph instead?
24.
Show HN: Icebug-format: immutable, interoperable graph standard(github.com)
3 points by adsharma 27d ago|0 comments
25.
by adsharma 29d ago
[ mojo explaining why they didn't create an embedded DSL] > This is particularly problematic if you're trying to introduce fundamental new concepts because you can't change the grammar of Python or C++. Yet, this is exactl
26.
by adsharma 29d ago
Mojo as MLIR++ is how I've thought about it as well. Many users will find transpiling static python to mojo an interesting path. Updating py2many --mojo to 1.0 to see what breaks. Also looking forward to translating existing python adt
27.
by adsharma 1mo ago
Given the shared heritage with MLIR, one way to think about mojo: higher level IR, but still an IR. Rpython for GPUs. But mojo is a superset, not a subset. So why not use a subset, infer what you need and generate mojo? I've never seen
28.
by adsharma 1mo ago
https://github.com/py2many/static-python-skill You can stick to python and generate mojo or rust or lean. There is more than one solution to the two language problem.
29.
by adsharma 1mo ago
If you must use Django, use it through an abstraction layer like this: https://adsharma.github.io/django-fquery/ Your models can be plain old python data classes, declaratively mapped to Django primitives.
30.
by adsharma 2mo ago
- Installing a db these days is not prohibitive. You don't have to hire a DBA curl <url> | bash - Some graph databases allow you to query a relational source without migrating data

More ›