DavidB

The limit of representation is intelligence |
Human? | Engineer | MD |
Ex @Grappaxyz @GoogleAI @Theteamatx @Verily @TechnionLive @HarvardHBS

[DavidB @DavidBensh]

@DanielMiessler "Continuous efficient adaptation with limited resources"
I couldn't find a better and more concise definition than that for intelligence.
All other definitions seem like an instantiation of it as the reward function.
(i.e compression is an adaptation to limited memory resources)

[DavidB @DavidBensh]

@mgostIH You can think of it as a compromise between token by token generation and diffusion.. where the later exploits fixed correlation "scaffold", this limits it to two directions which is more intuitive for thoughts, stories and linear reasoning.

[DavidB @DavidBensh]

@LordCoward_ @elonmusk @POTUS is staring at you helplessly while you wipe boogers on the Resolute Desk.. at age 3!
How can you take the rest of your life seriously after that? 🤨

[DavidB @DavidBensh]

@javilopen @360_Hardware "Loop style" consciousness is a prevalent philosophy. Though not many experiments are being done imo.

This is old but very original:
https://t.co/pYtyKoeFhu

There are fractal aspects to that framework which are also fascinating.
Read "I am a strange loop" if you havent already.

[DavidB @DavidBensh]

Attempts like this should be aimed at more than just "dry" explainability.
One theory is that our consciousness is basically an observer brain "watching" our acting brain and trying to make sense of inputs/outputs with inflated illusion of control..
This can get very interesting!

[DavidB @DavidBensh]

@AmitMandelAI @Tommyziv7 TensorFlow losing to PyTorch in the long run also marks a missed opportunity to enable an alternative SW framework..

[DavidB @DavidBensh]

Because we cannot see ourselves in the reflection of a genius, awe turns to detachment.

What will our sense of value become now? Will we penalize strokes of genius?

[DavidB @DavidBensh]

We’ve been looking for the wrong risks of AI, while one threat already taken root.

Extraordinary today provokes skepticism, as the belief grows that remarkable must be machine-made.

<1/2>

[DavidB @DavidBensh]

@elonmusk Does this mean Elon is POTUS?

[DavidB @DavidBensh]

@BeckettUnite @UNRWA was established to care for refugees of a war that took place more than half a century ago!
The fact that they did not reinstate a single Palestinian and prefer to teach children to stay put and "liberate their lands" should tell you everything you need to know!

[DavidB @DavidBensh]

@itamar_mar @gdb Consciousness is too abstract to define or test, I'll push that to the next sprint 😅
Though, incorporating semantic frameworks in autoregressive models like transformers is an active (and very exciting) research area..

[DavidB @DavidBensh]

@karpathy I asked it to code reversehash_SHA256.py and it didn't even return code.. these LLMs are useless!

[DavidB @DavidBensh]

@sama @mediaking no soup for you!!

[DavidB @DavidBensh]

@Yampeleg Some people would rather do 32 quadrillion FLOPs every second than actually going to therapy..

[DavidB @DavidBensh]

@tunguz In the old days they actually used TensorFlow behind the scenes so there is something to that..

[DavidB @DavidBensh]

@tunguz High dimensional euclidean spaces bring everyone closer..

[DavidB @DavidBensh]

@0xidanlevin Isn't that e/acc? @beffjezos

[DavidB @DavidBensh]

@GaryMarcus Eventually we'll get there with multiple language models, graph databases, bunch of alternative non-linearities and a fancy random number generator.
All these "token prediction is dumb" crusades would be very amusing in retrospect..

[DavidB @DavidBensh]

@siyan_zhao Is inference fully determintic on the LLM classifiers? (i.e most probable next token)

[DavidB @DavidBensh]

@TalMorgenstern Fellas from https://t.co/OqKL9dst7n and/or anyone trying to train/finetune Hebrew models for ASR/LM.