Ask HN: What tech is under the radar with all attention on ChatGPT etc.?
ChatGPT and other generative AI seems to be taking a lions share of mindspace in the tech industry right now.
I'm curious to hear what interesting new things people are seeing that AREN'T trendy right now (yet?!).
175 comments
[ 5.7 ms ] story [ 210 ms ] threadDisplay panel tech
Silicon photonics
now the display can directly serve ads!
(Like, firmware? Or is this more like an all-in-one type computer but the 'computer' part is equivalent to a Raspberry Pi/Arduino/etc [So - programmable, but somewhat minimal?]?)
I can't wait to get a 500 KWH battery backup for my house for $10k...eventually...
At first it sounded amazing but then when gazing at this number it struck me suddenly - isn't this like an energetic equivalent of 0,5 ton of TNT ?
So if I rephraze this statement to "I cannot wait when I will be able to store half a ton of TNT in my garage" this whole premise starts to sound less and less appealing.
(especially since there was a electric scooter battery explosion in flat near me which killed someone)
https://mlc.ai/web-llm/
Seems like the ecosystem is very early/non-existent.
The WSL in the article was throwing me off.
Matt Rickard also had a short write-up recently: https://matt-rickard.com/implementing-llms-in-the-browser
My take is that we probably won't see much training done on WebGPU, because training is done upfront and it makes sense to standardize on a single GPU vendor and use an interface that can squeeze all the juice out of those GPUs (CUDA). But for inference and run-time computations, it could be very interesting to take a model trained with CUDA/PyTorch and export it (maybe with Apache TVM or tensorflow.js) into WebGPU that can run on end-user devices.
In its current state, can you train on PyTorch, export to ONNX, load ONNX in JavaScript/WASM, then use it for WebGPU inference?
I'm not trying to sound obsessed/married to ONNX, I just though it was "the standard". Curious to learn alternatives/what people are doing now but I fear even talking about what might being done is discussing "bleeding edge".
Edit: A quick Google shows yes https://onnxruntime.ai/docs/tutorials/web/
I believe so. Onnxruntime very recently merged a WebGPU backend: https://news.ycombinator.com/item?id=35694553
You can also go directly from PyTorch to WebGPU with Apache TVM. (ONNX is also supported, but my understanding is that it's better to go direct). This is an example using an LLM trained with PyTorch (I think) and run in the browser: https://mlc.ai/web-llm/
https://github.com/microsoft/onnxruntime/releases I don't see it in any releases yet?
https://github.com/microsoft/onnxruntime/milestone/4 I don't see it in the upcoming milestone.
I don't see any examples or docs that go with it
https://github.com/microsoft/onnxruntime/wiki/Upcoming-Relea... This seems to be out of date
https://github.com/microsoft/onnxruntime/tree/rel-1.15.0 I do see the js/webgpu work merged into here so I guess it'll be released in 1.15.0
https://onnxruntime.ai/docs/reference/releases-servicing.htm...
> Official releases of ONNX Runtime are managed by the core ONNX Runtime team. A new release is published approximately every quarter, and the upcoming roadmap can be found here.
ONNX Runtime v1.14.0 was Feb 10th
As for people continuing to use three.js, yeah, I still see a lot of folks using it to build games. That whole “js everywhere” still seems to be really popular with a lot of folks.
It probably also helps that new comp sci students learning 3D graphics are often now taught using three.js instead of C++ like when I was a student.
Epic released Unreal Engine for Fortnite about one month ago.
Meta have some interesting stuff in their pipe.
I think that especially Roblox will shift how the younger generation view games, with more emphasis on social. It is their MMOs.
In the hardware space openBCI has partnered with Valve and they’re working on an BCI integrated immersive headset called Galea. It runs about $25k+ atm. They’ve begun shipping to researchers and developers and I’ve heard they’re aiming for a consumer line but don’t know of a planned release date.
Metaverse is merely premature. AI is the bigger and more general tech wave, but Metaverse will start to crest in 5 years or so.
Think about all the immersive, interactive fiction powered by LLMs and AI. That's just scratching the service of what's possible.
Games, movies, narrative, and social will converge.
The closest you could say was games have a narrative woven in. Movies are passive and everything is intentional. Games are fundamentally about interactivity. They are practically opposite forms of entertainment.
I don't even know what you mean by 'social will converge'.
Here's my prediction:
- Stories, narratives, characters, and game loops will become generative. Their capacity to generate new unseen and unwritten content will be heralded as revolutionary.
- The quality of games and GenAI will soon reach that of Hollywood films and take on a cinematic feeling.
- Players or viewers will be able to actively participate in the unfolding world (games) or sit back and watch them unfold (movies). This will be a spectrum (and not just on a single dimension).
- Narrative instances will be passive, single player, multi player, and massively multiplayer. You can sit back and watch a world unfold with your friends, or each participate as your own individual character that is an immersive and key part of the world. Cooperatively or in competition with one another, and at every point within that that spectrum. This will be similar to D&D, but at a scale never before seen. Wholly engrossing.
- Just like we've seen inklings of with YouTube and TikTok, you'll be able to remix the narratives you construct and participate in. You'll be able to world build, tweak, and share. You'll be able to take off the shelf situations and narratives and customize them as much as you want. Your previous "play through" or "stage time" can be recorded as its own personality that can become an AI agent in someone else's journey.
- This world building scene will be full of creators, musicians, directors, actors, and lots of people that just want to experience and hang out.
- Just as before, I'll throw porn into this prediction too. GenAI porn will similarly replace the manually filmed artifact, and moreover, it'll be interactive. Some people will even fall in love with their artificial companion(s). As this becomes more normalized, these adult AI companions may even find themselves injected into otherwise family friendly narratives. What you do with them at home is your own business.
Before you say it's implausible, I know for a fact that 3d-guided diffusion works and that it'll soon spook the entire Hollywood industry to the core.
[1] https://news.ycombinator.com/item?id=32711614
[2] https://news.ycombinator.com/item?id=35523656
[3] effectively I have. I quit my $400k/yr job to go full time on this and poured over half a million of my personal funds (so far) into building my company
No, I told you it wouldn't devastate an industry built on hosting videos. It hasn't and those video hosting sites already host videos that are animated, because they don't care how their stuff is generated if people want to watch it.
Go back and re-read what you linked with a little more focus this time.
poured over half a million of my personal funds (so far) into building my company
And how many people are paying for your 'generative game loops' (whatever that means)?
I'm morbidly curious: how do you plan to recoup your costs? I assume porn is so trivial to get for free that it must be a relatively low profit industry...
I always think of the oft-quoted meme about porn being the canary in the coal mine for the adoption of new tech: film, VHS, usenet, web, P2P, Twitter/Tumblr/Reddit, streaming, VR, AI... I think the trend has held pretty well historically.
NovelAI, Civit, etc. are making a lot of money on it, and they're leading indicators of more expansive "SFW" use cases that will follow.
I’ve considered quitting my job and pursuing it full-time. Being an early adopter here seems like a sure fire way to succeed. I’m curious what your journey has been. I’m launching my product next week. Have a single customer signed on with a plan to scale it out, should the results come back positive.
every 10 year old in N. America is on those and/or minecraft. no way the metaverse has that kind of reach anytime soon.
IT is cyclical.
killin' me smalls.
Only somebody who does not work at FAANG uses FAANGMULA, so why not just go with MULA.
Instead of renting a VM or running docker containers you rent a small physical server for each service.
Or of course world war 3 happens and the latest, greatest thing going in IT is “I mean I’m still alive, whatever that’s worth.” And making anything usable from whatever is found. Even a blinking LED becomes either neat or a way for a roving warboy to find/kill you.
"All of this has happened before, and it will all happen again." ;-)
That said, Lambda has been out for a while and seems to have avoided taking over the world. Generic vCPU-hours are so cheap now though that it isn't compelling from a a cost perspective (serverless will either cost you more at the high end of the scale, or else save you a few bucks a month on your idle instances at the low end). Also, the developer experience isn't as good yet IMO. Thinks like LocalStack help, but it's still not natural-feeling to deploy a big application this way. (scale-to-zero it's great for small side projects though -- I have a few apps in Cloud Run that cost me like a nickel a month, a few dollars if I have a good month in terms of traffic).
I do not know if this is solvable: a permanent service with light autoscaling handles even inconsistent load so well and reduces operational complexity so much that I don't know if serverless will be anything more than toys. It is no coincidence that AWS Lambda's first language was a frontend language* (Node.js). Those use cases have low to moderate scale (because after that you split frontend and backend).
^Amazon Prime Video post: https://www.primevideotech.com/video-streaming/scaling-up-th.... *First mention of AWS Lambda languages in AWS Compute Blog only mentioned nodejs: https://aws.amazon.com/blogs/compute/container-reuse-in-lamb...
The biggest problem is that the computing model, or the architecture that is commonly required for this sort of apps (event-driven and asynchronous) is hard to develop, manage and reason about (especially with the async-lambda integrations that by default retry their work if they fail). This means that you need to think about them and architect them in a way that makes them idempotent.
If you configure everything properly, event-driven, AWS Lambda-based apps can be extremely useful. I just don't believe that the complexity required to add them to your application is worth the complexity that needs to be added to your app to make it all work.
What I wanted to say is that Lambda functions have their use-cases, and that if used reasonably, they can heavily simplify the application that they are used within.
Our product (https://stacktape.com, disclosure: Im a founder), makes it easy to deploy both container-based (ECS Fargate and ECS EC2) apps, and lambda functions, with almost no changes in the configuration). This allows our users to always choose the right technology/computing model for the use-case they are working on.
What about 'AI Tendrils' <-- tiny AI workers on little devices that feed the Mother AI - like an agent would monitor, but you tell it what info or behaviours are allowed and have that securly streamed back to the AI/model/LLM? (I dont know where the data would go) but then when you run your query/prompt youre running it against a much closer-to-processing-power than if you had to go query each device on a request?
--
I thought this output from GPT was interesting for thinking of the above:
https://i.imgur.com/WsxwuJh.png
and a BOM
https://i.imgur.com/5g1mXfp.png
https://i.imgur.com/MdbQKx1.png
https://i.imgur.com/Bi0PRTQ.png
--
Actually, I asked gpt to recommend additional columns, here is the result
User what columns would you suggest to add ChatGPT
Here are some additional columns that you may consider adding to the HN table:
* Display type: OLED, LCD, AMOLED, etc.
* Display resolution: the number of pixels on the screen.
* Water resistance: the level of protection against water damage.
* Heart rate monitor: whether the wearable device has a built-in heart rate monitor.
* Sleep tracking: whether the wearable device can track your sleep patterns.
* Music playback: whether the wearable device can play music and if it has built-in storage.
* NFC: whether the wearable device supports near-field communication for contactless payments.
* Voice assistant: whether the wearable device has a built-in voice assistant like Siri, Google Assistant, or Alexa.
* Other sensors: such as an accelerometer, gyroscope, barometer, altimeter, or compass.
Keep in mind that the more columns you add, the wider the table becomes, which may make it more difficult to view or analyze. So, you may want to choose the most important columns for your specific use case and audience.
Also, somewhat related: secure anonymous data exchange has a lot of cool possible applications beyond health exposure tracking.
Most VR application/game developers limit their polygons/resources to that of a mid-range mobile phone because that's about what the stand-alone headsets can deliver in terms of graphics/memory performance. However, if they knew that end users were going to be running their applications with PC-level hardware it would enable a whole heck of a lot more detail, better AI, level generation, and all-around functionality in VR applications/games.
Which ones are you referring to? Are they complete and ready to ship products? Or is that a DIY project / kickstarter?
I thought about splurging for a full-color model but wasn't sure whether the tech was there yet. Do you think it's worth the upgrade or should I hold off a little longer?
But Apple's going to be launching their headset next month
I think your attitude about the matter could use some adjustment. There is more to it than age.
ChatGPT knows a lot about existing tech - even more than people who think they're pretty steeped in it (including myself here) - but inventive stuff that doesn't show up after a bit of googlefu, it's much less helpful (even useless).
So great assistant, great rhythm section but falls down as lead.
For now, anyway (and open question, for how long?)
Intrinsic AI making a product announcement in two weeks.
Deepmind collaborating with Robotis implementing deep RL on the robocup humanoids to play soccer at the next level.
https://github.com/cnlohr/ch32v003fun
Fully open source stack!
Grab the below eval board and peck around: https://www.aliexpress.com/item/1005005269690018.html
Also see here for awesome writeup on the BL602 chip using the Nuttx RTOS: https://lupyuen.github.io/articles/nuttx
It is the best ISA around, and it is an open standard with wide industry support.
>over something like an ESP32
Espressif has stated intend to transition to RISC-V entirely. Newer ESP32 like the C3 and C6 are already based on RISC-V.
https://www.helionenergy.com/ https://cfs.energy/ https://www.tokamakenergy.co.uk/ https://tae.com/
Insert below; jokes about its being 25 years away, and comments about how it will never work etc
Edit: Dang... It was just a joke. (well partially, the timescale should have made it obvious still)
My concerns have moved to "okay, what is the cost of things even if energy is 'free'?". For example, desalination -- currently the majority of the cost of desalination is in the energy used, but even if the energy is free, the materials cost and capital expense is currently about 40% of the cost.
Will the cost of the inputs also go down dramatically as a result of fusion? Or will desalination still be cost prohibitive for most developing areas due to materials? If desalination price falls 99.9% -- what environmental costs will be associated with the extraction/transportation of necessary materials as well as the warm & highly saline effluent that will be dumped into surface waters?
With unlimited clean energy, some ridiculously inefficient schemes can become practical, such as removing CO2 from the atmosphere.
(writing from Sweden where residential geothermal heat pumps were installed more than 20 years ago even in our cold climate)
Is probably the most "under the radar" of all. Is interesting how little attention you get from people about this stuff (when talking about implementation or internals).
Is like, literally, only make a splash AFTER at-minimun you have something competitive with a major stablished product.
But you get more "views" talking about compilers than talking about DBs!
(except when talking about compilers that use the relational model. I probably should rebrand https://tablam.org as a "lisp"! -no accurate, but whatever-)
* I also pithy anyone doing a new OS, now that is crickets!
2. Bambu Labs leapfrogged prosumer 3D printing by 5-10 years. Their printers are becoming the de-facto standard. This will enable all sorts of startups for the next decade.
3. Most iPhones now have a 3D scanner + AI-aided software to accurately scan your body and generate designs is becoming more mature. See also 2.
4. material sciences both new discoveries and new applications for 50 years old discoveries
I'm not very well versed in the subject but this should point you in the right direction!
It seems they’re in a unique position because sugar cane grows there, is used in industry, and the leftovers that aren’t usable for anything else can be turned into fuel.
I had no idea Brazil has a unique bio ethanol industry, thanks for sharing.