John Platt (@johnplattml) header
John Platt (@johnplattml) profile picture
Seattle, Washington
Joined June 2014
Photo by John Platt (@johnplattml)Photo by John Platt (@johnplattml)Photo by John Platt (@johnplattml)Photo by John Platt (@johnplattml)Photo by John Platt (@johnplattml)Photo by John Platt (@johnplattml)Photo by John Platt (@johnplattml)Photo by John Platt (@johnplattml)Photo by John Platt (@johnplattml)Photo by John Platt (@johnplattml)
Electricity use at data centers did NOT skyrocket. In fact, despite massive growth in computation and data storage provided, total electricity use barely budged. #EfficiencyFTW
This Post is from an account that no longer exists. Learn more
5
16
1
83
54,186
✨ Today we're launching AI coding features to our Pro+ subscribers located in the United States, including natural language to code generation, code completion, and an integrated chatbot. We'd love to hear from you with positive examples of AI coding in Colab: please share! 1/
14
138
33
769
170,833
🥳 We've launched support for Google children's accounts! Children under the age of 13 are now able to access Colab for learning Python programming and machine learning. 🧐
14
149
27
628
138,664
Is Elden Ring an existential risk to humanity? That is the question I consider in my new blog post. Yes, it is a comment on the superintelligence/x-risk debate. Read the full thing here: togelius.blogspot.com/2023/0… Or follow along for a few tweets first, if you prefer.
32
133
50
522
496,907
I got really excited that the LLaMA paper calculates and reports their carbon footprint! 🦙🌬️🌎 But upon looking at the paper itself, it has this table, which completely misconstrues the emissions of OPT and BLOOM, while not actually reporting LLaMA's own. How? A thread 🧵
14
63
14
290
250,646
John Platt retweeted
We’re proud to share promising new research published in @SpringerNature’s @NatureComms journal. Along with our partners at @NIFSplasma in Japan, TAE has conducted the first-ever hydrogen-boron fusion experiments in a magnetically confined fusion plasma. rdcu.be/c52Ym
1
19
5
48
8,255
For DECADES we've assumed the surface code uses 4 couplers per qubit. But change perspective a bit, and it's clear it works just as well with 3 couplers per qubit. Same distance, same # of gate layers, same footprint. Also 23 other circuit variations: arxiv.org/abs/2302.02192
Forget what you know about surface code circuits: arxiv.org/abs/2302.02192 Our new work harnesses the time-dynamics of circuits to do the surface code on a hex-grid, use ISWAP gates, and move the patch. These circuits relax hardware requirements without reducing performance. 1/9
2
6
44
9,763
People are asking if there are alternatives to RL in RLHF. Yes, imitation with DAGGER (tutorial: ri.cmu.edu/publications/an-i… ). The user provides feedback with corrections, e.g. when the agent says “that” the user tells the agent that instead of saying “that”, it should say “this”.
This. People are perhaps too quick to attribute favorable results to the RL piece of this formula.
2
15
2
102
52,306
John Platt retweeted
Transformers are all the rage today. But neither DALLE nor Stable Diffusion uses Transformer for image generation. Instead, they rely on a 7-year-old, Jurassic-era neural architecture. Why? 🤷🏾‍♀️⁉️ It’s finally time for Transformer and Diffusion to join forces! Quick 🧵👇:
23
227
29
1,434
339,188
Introducing RT-1, a robotic model that can execute over 700 instructions in the real world at 97% success rate! Generalizes to new tasks✅ Robust to new environments and objects✅ Fast inference for real time control✅ Can absorb multi-robot data✅ Powers SayCan✅ 🧵👇
57
516
128
2,256
Sharing with you this evening, video from #MaunaLoa's NE Rift Zone. At ~11 am on Nov 29, 2022, the lava fountains were measured at 35-40 meters (115-148 ft) in height. The second video shows an ‘a‘ā lava flow several meters (yards) thick, moving to the northeast. #MaunaLoaErupts
26
417
59
1,312
New paper in Science today on playing the classic negotiation game "Diplomacy" at a human level, by connecting language models with strategic reasoning! Our agent engages in intense and lengthy dialogues to persuade other players to follow its plans. This was really hard! 1/5
74
717
257
3,672
Yeah, bt what if we cld have a ~70 node, ~210 line UG HVDC trans. overlay for CONUS? Cost ~$400 bn (or $35 bn/yr or $7.5/MWh). It wld save 4.5 Gt of GHG emissions by 2050 & enable reduced electricity costs of $25/MWh. #climatechange #cleanenergy #UGHVDC #justsaying #permitting
8
15
4
91
We sequence SARS-CoV-2 from wastewater using a different approach than most. We sequence smallish fragments of the genome so that we can tell if particular mutations are derived from the same virus. This way we are better able to determine if there is anything new circulating.
1
21
1
195
John Platt retweeted
// Stable Diffusion, Explained // You've seen the Stable Diffusion AI art all over Twitter. But how does Stable Diffusion _work_? A thread explaining diffusion models, latent space representations, and context injection: 1/15
91
1,098
148
4,851
John Platt retweeted
Learn how we combined our latest language model, PaLM, with robot learning algorithms to create PaLM-SayCan, a robotics system that uses natural language to complete complex tasks in a real-world environment → goo.gle/3QRJhgl
5
130
25
386
Excited to launch a Quantum Virtual Machine which emulates the experience and results of programming one of our quantum computers. It will make it easier for researchers to prototype new algorithms and help students learn how to program a quantum computer. blog.google/technology/resea…
80
405
49
2,386
As someone who has spent easily half a decade staring at AI arXiv each week and trying to articulate rate of progress, I still don't think people understand how rapidly the field is advancing. Benchmarks are becoming saturated at ever increasing rates.
49
352
60
2,336
We now finally have the thing I've always wanted: an analysis of the best vision models for finetuning. In joint work with @capetorch we tried nearly 100 models on two very different datasets across various hyperparams. Here's the top 15 on the IIT-Pet dataset: 1/🧵
13
149
7
645