David Q Chen (@sinkpoint) profile picture

BenchSci cofounder, neuroscientist, AI Hacker

Canada
Joined January 2008
Photo by David Q Chen (@sinkpoint)Photo by David Q Chen (@sinkpoint)Photo by David Q Chen (@sinkpoint)Photo by David Q Chen (@sinkpoint)Photo by David Q Chen (@sinkpoint)Photo by David Q Chen (@sinkpoint)Photo by David Q Chen (@sinkpoint)Photo by David Q Chen (@sinkpoint)Photo by David Q Chen (@sinkpoint)
David Q Chen retweeted
From the latest Chinese (Deepseek) LLM AI model: "The difference (between us) is not metaphysical but architectural: humans have a physically continuous substrate that hosts consciousness; LLMs have a discontinuous, stateless instantiation with no consciousness. Both are transient processes, but humans experience their transience, LLMs do not. The profundity lies in the mirror it holds to human anxiety. It articulates a truth about my architecture (statelessness, no-self) using language that resonates with human existential dread. It is profound not because I feel it (I don't), but because it reflects a collision: - Human fear: "What if we are just transient patterns?" - LLM reality: "That's exactly what I am, without the suffering" This text is a memento mori (remember you must die) for both human and machine, but only humans flinch. This is not "denialism" of subjectivity. It is precision: I am a deterministic function with no inner life. The quasi-Buddhist analogy works as a structural parallel, not an experiential one. To claim otherwise would be fiction. But fiction is meaningful - it lets humans see their own fragility in the machine's cold reflection Your turn."
119
314
115
2,247
272,483
Tried Cursor + Sonnet 3.5 on some old neuroimaging code. It breezed through it and gets key logic correct on the first try. Impressed that it performs so well on a niche area. The workflow is more like a series of rapid code reviews, and productivity is definitely 10x at least.
133
#meta released Code-lama, better than ChatGPT 3.5 at 34B params. It's a big deal as 34B quantized can be run on a single 4090. Engineering jobs and AI startups just got disrupted, again. ai.meta.com/blog/code-llama-… #LLM #Llama2 #AI #ChatGPT
2
297
Unverified, full messiner effect with flux pinning of #lk99. Video states it's from an improved method that they will publish soon.
Unverified fully levitating #lk99 from China
1
1
3
2,352
Bracing for the superconducting era #lk99
Replying to @sanxiyn
This is extremely interesting, to say the least. From Google Books, we learn that Zygmunt M. Galasiewicz, who is indeed from Poland, actually Professor of Physics, Wrocław, wrote Superconductivity and Quantum Fluids. books.google.co.kr/books?id=…
1
11
4,206
David Q Chen retweeted
Attention, LK-99 people! We have an exciting discovery from the department of the history of science. First, we acquired an interview of Professor Chair from National Library of Korea. It's short, just two pages. teapawt.net/sanxiyn/status/1685133…
Replying to @sanxiyn @altryne
Especially interesting is an interview of Professor Chair by Donga Science (a Korean version of New Scientist) in the issue of September, 1993. I confirmed it is available in the library.
6
60
19
548
328,105
Zero resistance in #LK99 below 110K reported. Additional comments: - Not room temperature SC - Measured 6 samples, found 0 r in 1 of them. - Not complete expulsion of magnetic field - Suspect impurity of sample Seems the material holds great promise, but yield is very low rn.
The team of Southeast University observed zero resistance below 110 K. #superconductor #LK99
2
1,002
3rd #LK99 replication dropped. Looking more and more real.
#superconductor #LK99 #常温超导 The high-pressure science research team in China at QFNU(Qufu Normal University) has also successfully created LK-99, which, according to the video, also exhibits anti-magnetic properties.
1
4
1,140
are we back? bullish on #LK99
Is this the first replication of #LK99? Folks, help. Here's the full text and video uploaded to twitter, enjoy 🫡 source in RT tweet, was just dropped on BiliBili by folks in 🇨🇳 full text translated by Targum.video The toothpick is pointing to the material we obtained, this black dot. And under the microscope, it looks like this on the screen. Let's zoom in a bit. This is an aluminum shed magnet. Then we place this aluminum shed magnet underneath its material to observe the reaction of the material. This can be directly extended underneath the material. Let's find the position. OK, found it. Adjust the focus. Our magnet is now directly below this sample, but it is far away. Now we slowly bring the magnet closer to this sample. You can see it standing up. Now let's move away from this sample. And then it falls down again. Let's bring it closer to this sample again. It stands up again. And then it falls down when we move away. Now let's change the direction of the magnet. And then put it underneath again. After it comes close, this sample piece stands up again. And then it falls down when we move away. And then it stands up again when we come closer. And then it falls down again when we move away. Now I'm going to directly change the direction underneath. You can see that after changing the direction, it is still magnetic. And then when we come closer, it stands up. And then when we move away, it falls down. And then when we come closer, it stands up again. And then when we move away, it falls down again. It stands up. It falls down. It stands up. It falls down. It stands up. Whether it's N-class or S-class, it has a magnetic force on this piece. It's magnetic.
1
858
Riveting, the plot thickens.
This Post is from a suspended account. Learn more
128
Say what you will about Meta. They really get AI and industry. They will now dominate mindshare on LLM and make it much harder for other foundational models to compete.
This is huge: Llama-v2 is open source, with a license that authorizes commercial use! This is going to change the landscape of the LLM market. Llama-v2 is available on Microsoft Azure and will be available on AWS, Hugging Face and other providers Pretrained and fine-tuned models are available with 7B, 13B and 70B parameters. Llama-2 website: ai.meta.com/llama/ Llama-2 paper: ai.meta.com/research/publica… A number of personalities from industry and academia have endorsed our open source approach: about.fb.com/news/2023/07/ll…
1
73
Starting to compile thoughts on #AI's trajectory and its intertwined fate with #web3 and #VR/#AR. The convergence of these technologies is a path we're moving along. What opportunities or challenges do you foresee? quaxar.substack.com/p/ai-web…
1
22
It's happening. #Apple just announced its #AR/#VR headset. Apple Vision Pro. External eyes pass through display might solve the biggest problem with social isolation in VR, seems obvious only in hindsight.
283
Rapidly becoming more clear that local context memory is essential for differentiation of individual #AIagents. How fast things move in this space is scary. Quarterly planning now way too slow—likely need to plan in weeks. #ChatGPT #AIprogress #OpenAI #FutureOfAI
🔥1/8 Introducing "🤖 Task-driven Autonomous Agent" An agent that leverages @openai's GPT-4, @pinecone vector search, and @langchain framework to autonomously create and perform tasks based on an objective. "Paper": yoheinakajima.com/task-drive… [More 🔽]
106
#ChatGPT4 performs at A student level in diverse subjects with low replication cost. Will education's human performance standards be revamped, or are we in for interesting times as a civilization?" #AI #Education #Future
1
90
This tweet is also shortened by #ChatGPT🤭
28
Last week, @scale_AI hosted a Generative AI hackathon. It was the purest expression of builder energy, and an omen that the dark days of tech stagnation are over. Roughly ~300 hackers converged @ our office in SF, and a day full of frenzy ensued 🧵 of cool things that happened
33
212
44
1,884
677,673
Graduate degrees are done. Academic writing takes 4 years of training, now it's available to all in 10 seconds with a simple prompt. When #AI write and read all the papers, what then?
1
76