I think both can be true - ie. We can harness energy from the sun at much bigger scale than nature has so far, yet we can still learn a lot from how nature has spent massive parallel search over millions of years on coming up with incredible nano engineering we still have no clue how to do
I think Jeff Dean is right in that we will see much more specialised silicon in the future.
If something more bio inspired ie. predictive coding and in-memory compute fundamentally makes continual learning and much lower energy consumption possible there will be specialised hardware for it at some point
FWIW I think the brain has multiple “learning rules” and operates at multiple timescales
Which era Trane are you talking about? The man went through at least 4 major shifts. I’m guessing you are talking about Giant Steps and the later avant garde years?
He did a lot of stuff too which is not dense at all and very approachable - Naima, A Love Supreme, My Favorite Things, Kind of Blue with Miles, a lot more, enjoyable, classic albums.
It's all subjective. The idea that none of these are dense is pretty bonkers to me. Some people don't find hard-bop to be approachable. Kind of Blue is really more the exception to his typical output.
Plus you get the license for free when you buy a camera from them!
I’ve been using DaVinci Resolve Studio for 12 years now with frequent updates, with the license I got with the original Pocket Cinema Camera for £800 (still a fantastic pocket size cinema camera, shoots raw cinemadng and ProRes with a beautiful sensor).
Yes the voice part of OpenAI realtime/voice mode is great but it’s pretty dumb compared to newer models and often gets stuck repeating itself.
Google’s Gemini flash live 3.1 is better, especially used via the API - it can do tool calling (including to other, even smarter LLMs if you set it up yourself), you can set the reasoning level (even high is still close enough to realtime) and it can ground answers in google search. I love bidirectional voice and right now it’s probably the best option. You can try it in AI studio
Give it a shot, 3.1 live one in AI studio/API and max out reasoning - not the one in Gemini app it’s an older model.
Another option is to use pipecat with their VAD and separate STT and TTS and any (fast) LLM of your choice - but it’s more plumbing and not a true speech to speech model
Haha, wow, I never thought I'd see a voice model that was too quick, but 3.1 live felt like it responded unnaturally quickly! I'm kind of blown away, I'd want to insert a 100ms delay to make it sound more natural, wow. I never thought I'd see that.
> Give it a shot, 3.1 live one in AI studio/API and max out reasoning - not the one in Gemini app it’s an older model.
Do you know why this is a thing? Despite the app technically being Gemini, I find it quite crap, while the AI Studio thing with thinking is my favorite LLM. Very jarring tbh.
I mean yeah FPAA’s would be awesome and I used to wish for something like it coming from a discrete analog hobby electronics.
But in a my short two years in Analog IC design industry, i have been so divorced from the actual silicon that I rarely got a chance yet to go in lab and probe around the teeny tiny block I worked on in the complex labyrinth of the SoC. I don’t wish for it (I learnt the hard way, be careful what you wish for; and in this case, if I am in lab debugging something in silicon, means something terrible has happened to what I worked on and it might have cost the company about $200k or more), but someday soon i will get into the lab just to play around with the fancy ass oscilloscope.
In the meanwhile, I did realize the invaluable power of having a python frontend API for querying basic details of your devices. (Python and not SKILL/Lisp since it pretty much works with any AI, and is very well worked on) and AI has been okayish with it. I feel AI would be a good aid in actual circuit design if it understood the Topology of the circuit, which at this point I am tempted to say might require something akin to AST but for SPICE. However, AI has been awesome at regexes and scripting which is also the meh and boring part of the circuit design process.
The AST idea for spice is something i've thought about too. a netlist is already a graph, the LLM just can't see it that way when it's flat text. serializing it with topology intact, adjacency, port polarities, device semantics is basically what your python frontend is doing implicitly, which explains why it behaves so much better than dumping a raw netlist into the prompt.
How is it they can’t either go to Wikipedia or one of the LLMs (despite hallucinations, tend to get simple things right) and get some corroborating evidence before making such basic mistakes on an article?
Man I can’t even trust simple things these days from LLM’s. Hardly scientific but I just decided to do my own little test one time when I was on discord talking to some friends about The Game Awards back in December or so. ChatGPT would simply omit winners and/or categories - got it wrong (twice the same way, one unique way) 3 times. We tried Gemini, it gave 1 wrong answer and omitted 2 categories. It was impressive how much worse than a basic search they were at a simple “what were the results of the 2025 Game Awards?”
Easy install, discord/whatsapp/tg out of the box. And some agent orchestration out of the box where the main LLM can farm out tasks to different models/agents - yes Claude code has some of this too but I think this has more
