Dileep George
@dileeplearning
Followers
13K
Following
12K
Media
752
Statuses
6K
AGI research @DeepMind. Ex cofounder & CTO @vicariousai (acqd by Alphabet) and @Numenta. Triply EE (BTech IIT-Mumbai, MS&PhD Stanford). #AGIComics
San Francisco, CA
Joined June 2017
And here's the dirigibles analogy:
blog.dileeplearning.com
Notes for navigating large language models and beyond...
2
3
20
This discusion thread is really nice - Deleep always forces one to think more and provides creative logical additions
AI Consciousness, qualia, and personhood...my current thoughts. Can AI systems have consciousness? Yes, I think it is possible to build AI systems to have consciousness. While we haven’t pinned down exactly what it means, we will. Consciousness is related to information
0
1
3
Is the "New Math" behind the AI trade a fraud? Let me break down the three pillars of the AI economy and why even bulls like me are getting concerned about the numbers 🧵
1
0
1
AI Consciousness, qualia, and personhood...my current thoughts. Can AI systems have consciousness? Yes, I think it is possible to build AI systems to have consciousness. While we haven’t pinned down exactly what it means, we will. Consciousness is related to information
48
11
74
A note on AI consciousness, qualia, personhood....etc. https://t.co/2iish1hxNO
blog.dileeplearning.com
A note on my current positions, in a FAQ format.
0
0
0
I think @KordingLab and @eboyden3 got this wrong. We can simulate microprocessors because we understand the abstractions that connect the components to overall function…..not the other way around.
Applying new tools to entire brains, starting with C. elegans, offers the opportunity to uncover how molecules work together to generate neural physiology, and how neurons generate behavior, write @eboyden3 and @koerding. https://t.co/TMjnGwbAPh
5
17
140
10 out of 10. Would recommend. BillionToOne is a wonderful example of what's possible when you take on the most important, hardest problem you can. Oguzhan and David didn’t set out to “start a startup.” They wanted to solve a globally important problem -- and building a company
13
9
121
Our TPUs are headed to space! Inspired by our history of moonshots, from quantum computing to autonomous driving, Project Suncatcher is exploring how we could one day build scalable ML compute systems in space, harnessing more of the sun’s power (which emits more power than 100
824
2K
17K
A large fraction of interpretability work that interprets very broadly to produce dramatic narratives rather than grounded science certainly can't help with this. ["The model deceives us," "we have discovered the fear neuron," "this subspace encodes and plans deception!"]
I really genuinely believed that the deeper we got into interpretability, the more demystified these models would become and that the tendency towards anthropomorphization would fade away. I still don't understand why that doesn't seem to be happening.
3
4
50
I'm reminded that "functional networks" in neuroscience mean networks that are not going to work at all. (Not a joke, this is what it means). I guess "functional form" here is in a similar vein?
New Anthropic research: Signs of introspection in LLMs. Can language models recognize their own internal thoughts? Or do they just make up plausible answers when asked about them? We found evidence for genuine—though limited—introspective capabilities in Claude.
1
0
11
It was a pleasure joining @furrier, founder of SiliconANGLE, on the floor of the NYSE to discuss the future of finance on Solana. Our Chairman @KyleSamani was asked about Internet Capital Markets and its global potential
2
3
27
"you can't version-lock intelligence that doesn't actually **understand** what it is doing". LLM s don't *understand* the world like we do. What are the real ingredients of understanding? https://t.co/LRGkhhvpfZ
Gen AI is what happens when you ship something about 8 years too early and hope it doesn't catch up with you.
2
5
27
Friday evening with an injured foot....#AGIComics is how I'm coping with plantar fasciitis 😬 https://t.co/C486Tm6s5v
agicomics.net
A comic series.
0
0
1
Super excited for OpenAI to supporting growth of AI safety ecosystem -- in this case bio-defense.
One of the greatest opportunities in AI safety and security is the chance to help support the creation of new industry verticals that help maximize the benefits and minimize the risks. The technology is moving fast. One of the best ways to keep up is with more technology, more
12
13
148
this is not a blog...this is investigative reporting.
Revisiting last week’s open problems scandal, I wrote about LLMs as Lore Laundering Machines and why some are blind to the novelty whitewashing.
1
1
29
‘Intelligence is not about knowing! It is about knowing how to know! Measuring intelligence should be about measuring what the model knows about how to know‘
The one word in @RichardSSutton's Bitter Lesson that misled silicon valley. ----- A large portion of silicon valley misunderstood the Bitter Lesson essay to think that LLMs are bitter-lesson-pilled. I won't blame them. Some poor word choices in the essay give that impression.
0
1
3
A New Dawn rises. Supernova marks the next evolution of the network. Engineered for speed, scale, and true decentralization. Five years ago, the first Battle of Nodes tested the limits before mainnet launch. An epic competition still remembered today. Now, the competition
37
176
506
The one word in @RichardSSutton's Bitter Lesson that misled silicon valley. ----- A large portion of silicon valley misunderstood the Bitter Lesson essay to think that LLMs are bitter-lesson-pilled. I won't blame them. Some poor word choices in the essay give that impression.
.@RichardSSutton’s Bitter Lesson essay is popular in Silicon Valley, for reasons Rich didn’t intend. Why? One poor word choice in Rich’s essay!! Had he used a different word in one place in the essay the world wouldn’t have thought LLMs are bitter-lesson pilled. Want to
12
15
106
I have an English text string to send. Is DeepSeek-OCR’s “10x compression” saying that by converting the text string to an image and compressing the image and sending it is 10x more efficient (at 3% error rate) compared to a lossy text compression method that would achieve the
3
0
7
Get to the face of the clot with Carrier Delivery Catheter: new data reports Carrier delivered the aspiration catheter to the face of the clot in 100% of cases! Choose consistency, choose Carrier. Check out the article here.
0
5
22
.@RichardSSutton’s Bitter Lesson essay is popular in Silicon Valley, for reasons Rich didn’t intend. Why? One poor word choice in Rich’s essay!! Had he used a different word in one place in the essay the world wouldn’t have thought LLMs are bitter-lesson pilled. Want to
12
2
29
This is a fun analogy :)
And here's the dirigibles analogy:
0
1
2