Gottfried @gottfriedmath

Joined April 2017

Tweets

662
Followers

40
Following

49
Likes

1K

Gottfried @gottfriedmath

2 years ago

The 3 major risks of LLMs are : 1. Not to use LLMs, 2. Blindly trusting LLMs, 3. Overlooking content production.

0 0 1 59 0

Gottfried @gottfriedmath

2 years ago

⁦@yudapearl⁩ . Note that I did not influence or nudge the answering LLM in any way. You came out spontaneously as reference #1 in the field. Deserved! perplexity.ai/search/29f579e…

0 0 0 5K 0

Gottfried @gottfriedmath

2 years ago

“Yann LeCun, Chomsky & Marcus will keep gaslighting on ChatGPT right up until the AI Singularity” by Paul Pallaghy, PhD link.medium.com/tKZD4QAeozb ⁦@artistexyz⁩

2 0 0 1K 0

Download Image

If you want to find deep thoughts about causality, read Lee Smolin amzn.eu/1yRUoJw. If you don’t trust me, start with the last chapter, before the epilogue. @artistexyz @yudapearl @soboleffspaces #Causalinference

2 1 2 213 1

Gottfried @gottfriedmath

2 years ago

Une belle illustration du caractère infini de l’invariance d’échelle fractale.

Kamoshika @kamoshika_vrc

3 years ago

Une belle illustration du caractère infini de l’invariance d’échelle fractale.

5 181 1K 91K 107

Download Video

0 0 1 56 0

Gottfried @gottfriedmath

3 years ago

Events are colliders in a causal DAG. Pure mediators do not add information to their single cause (timeless photons). Time is made of events. Each collider DAG is what gives time an arrow. Reversed it becomes a tree DAG which is not Markov equivalent. @artistexyz @yudapearl

1 0 2 65 1

Gottfried @gottfriedmath

3 years ago

1. In tabular data, features have no special order. 2. In visual data, pixels have spatial context. 3. In natural language, data have sequential order which partially overlaps and conveys causal order. 2 → CNNs’ successes. 3 → NLP transformers’ impressive achievements.

0 0 1 0 2

Gottfried @gottfriedmath

3 years ago

“A Survey of Causal Inference Applications at Netflix” by Netflix Technology Blog link.medium.com/bjWPs0DReqb

0 0 1 0 2

Download Image

Gottfried @gottfriedmath

3 years ago

Agency, the key to AGI, because observational data can’t suffice to get causal understanding. ⁦@yudapearl⁩ ⁦@artistexyz⁩ link.medium.com/3mKTkH1fbqb ⁦@soboleffspaces⁩

2 1 2 0 1

Download Image

Gottfried @gottfriedmath

3 years ago

medium.com/@gottfried.mat…

0 0 0 0 0

Download Image

Gottfried @gottfriedmath

3 years ago

medium.com/@gottfried.mat… @yudapearl @artistexyz @soboleffspaces

0 0 2 0 1

Gottfried @gottfriedmath

3 years ago

In a reversed causal DAG confounders become colliders and vice versa, correlations are blocked while others open up. Modeled with a cBN, probability distributions change and no longer fit the same observations. Causality supports time irreversibility. @yudapearl @artistexyz

1 0 4 0 2

Gottfried @gottfriedmath

4 years ago

Code is how algorithms become understandable by machines. Low/no code is how algorithms become understandable by humans. #alteryx #NoCode #rapidminer #orangebiolab #knime

0 1 3 0 0

Download Image

Gottfried @gottfriedmath

4 years ago

One can always replace a causal model X → Y with a model X → Z ← Y while conditioning on Z. This means that with observation only one may hold that causation is an illusion due to selection biases. Generalized, this view leads to a multiverse theory. @yudapearl @artistexyz