Big Data & ML
Joined January 2009
- Tweets 11,726
- Following 3,814
- Followers 819
- Likes 108,663
2,516 Photos and videos
Pinned Tweet
May 19
My latest Deep Dive, published in Towards Data Science: Introduction to Lean for Programmers.
I couldn’t find an introduction to Lean for experienced developers so I wrote my own.
Combined with AI, it’s one of the most powerful tools you can try now.
towardsdatascience.com/intro…
1
145
Ronen Lahat retweeted
May 27
Maybe you all couldn’t tell the difference between normal conscious experience and a 5th dimensional entity sharding 1/300th of your consciousness across 300 brains before calling all_reduce but I would definitely notice
12
7
171
9,391
May 13
A surprising childhood memory that was fake. There are currently no church bells in the Sagrada Família, it’s a recording played through speakers:
sagradafamilia.org/-/el-cari…
1
65
Ronen Lahat retweeted
May 12
Plot twist: this is how the ancient one sees the world, the world as a stage and we’re actors / puppets
2
9
1,216
45,654
Ronen Lahat retweeted
May 9
Agentic coding is a form of machine learning. Generated code is best treated as a blackbox artifact whose behavior and generalization should be managed via empirical evaluation, like with any ML model.
Feb 19
Sufficiently advanced agentic coding is essentially machine learning: the engineer sets up the optimization goal as well as some constraints on the search space (the spec and its tests), then an optimization process (coding agents) iterates until the goal is reached.
The result is a blackbox model (the generated codebase): an artifact that performs the task, that you deploy without ever inspecting its internal logic, just as we ignore individual weights in a neural network.
This implies that all classic issues encountered in ML will soon become problems for agentic coding: overfitting to the spec, Clever Hans shortcuts that don't generalize outside the tests, data leakage, concept drift, etc.
I would also ask: what will be the Keras of agentic coding? What will be the optimal set of high-level abstractions that allow humans to steer codebase 'training' with minimal cognitive overhead?
87
99
1,094
99,640
Ronen Lahat retweeted
Apr 26
Before the agents became reliably autonomous, we chatted with them in a terminal.
Their context could only fit a million tokens (max). They were slow. They had “hallucinations”. We inspected their every thought. We watched their every output.
The mid-20s were weird for us all.
22
78
1,953
406,079
Ronen Lahat retweeted
Apr 26
An LLM is as conscious as a rock, but rocks are conscious, everything is conscious, consciousness is the basis of reality.
1
125
Apr 17
You don’t need to wait for the Leidenfrost effect to fry eggs on a stainless steel pan. Influencers do it for the show. It will be too hot for eggs.
93
Ronen Lahat retweeted
public speaking skills 10/10
170
11,352
47,226
4,157,482
Ronen Lahat retweeted
Mar 16
First they take the dingle bop and they smooth it out with a bunch of schleem. The schleem is then repurposed for later batches.
They take the dingle bop and they push it through the grumbo, where the fleeb is rubbed against it. It's important that the fleeb is rubbed, becasue the fleeb has all the fleeb juice.
Then, a schlami shows up, and he rubs it and spits on it.
They cut the fleeb. There's several hizzards in the way.
The blamfs rub against the chumbles, and the plubis and grumbo are shaved away.
That leaves you with a regular old plumbus.
45
104
4,900
503,357
Ronen Lahat retweeted
Mar 16
Learning Alpeh Bet was always part of Jewish children education, these are doodled scraps from a thousand years ago, found in Egypt
1
19
1,597
Mar 8
This movie spiritually foreshadowed the Iran strikes and its message was “America is back baby”
1
186