Flux159

Yeah the reconciler docs were even worse when I was working with it, I had to ask the team to help with better docs or code pointers.

Makes sense to keep the boundaries between the different libs, but was just mentioning that single repo makes dev work much simpler from my experience (lib boundaries still enforceable with CI or tooling), but obviously there's a bunch of work to get that setup too so might not be super important at the moment.

Looking forward to see where this goes! Also, I think using an LLM to generate components from specs is an interesting idea as you mentioned in a different comment is a great idea to try out too.

Nice, it's an interesting concept and always like exploring new ways of using react in other contexts (react-three-fiber being easy to use for simple 3d projects, I have also written an Unreal Engine reconciler prototype earlier built on an Unreal JS plugin, but decided not to pursue it further).

Do you see the future of this project being something like a library of tscircuit components that you can compose together writing code rather than having to use Eagle or other pcb design software? Or I guess somehow autogenerating tscircuit components based on specsheets?

Looking at the code, I see that there's a reconciler here https://github.com/tscircuit/react-fiber/blob/main/src/lib/r..., but that seems to just be a thin wrapper to https://github.com/tscircuit/builder which is doing the heavy lifting of getting to a serializable format. One thing I'm wondering is why is there a need for separate repos since I feel like the tsx portion, reconciler, and builder are kinda needed together?

This is nice to foster some competition in hardware for model training, but the availability of these machines seems very limited - I don't think there's any major cloud provider allowing per hour rental of Gaudi2 VMs and Intel's own site directs you to buy an 8x GPU provisioned server from Supermicro for more than 40k USD. Availability and software stack is still heavily in Nvidia's favor right now, but maybe by the end of the year that will start changing.

I wonder how this compares to LGM outputs which also came out recently - https://github.com/3DTopia/LGM - specifically LGM looks like it's generating the additional viewpoints from text or an image and then creating a mesh whereas GaussianObject seems to require 4 viewpoints.

We're getting closer to workflows that can generate good 3d objects and texture them with diffusion models, there's a bit of work in integrating these into existing tooling though - also, animations and rigging for characters still seems to be a big issue (at least I haven't seen any good 3d generation models that can create rigged characters yet).

Have been using Sapling for a year and also used Meta's internal Mercurial fork before that - the interop with Github PRs and familiarity with branchless development is one of the huge benefits of Sapling.

Being able to just "sl clone <github_url>" and then use sapling CLI commands is nice, but also Sapling's interactive smartlog in VSCode https://sapling-scm.com/docs/addons/vscode/ makes it easy to learn commands as well while giving a UI to commit, amend, and move diffs.

I think the only thing Sapling's really missing is git-lfs support which would make repos with binary assets easier to manage.

Love seeing this with the code open sourced - also references another great procedural generation tutorial in http://www-cs-students.stanford.edu/~amitp/game-programming/...

I’ve used both Sapling cli & Graphite with github repos without having to pay anything.

They both have extensions for vscode as well that makes it easier to manage stacks.

I think that good tooling is a necessity to do trunk based development well - not just CI/CD and testing on PRs, but also being able to have stacks of commits in multiple PRs before landing as a whole (from ghstack, Sapling, or Graphite).

Having worked at Facebook before where managing stacks of diffs was far better than GitHub, I wish that more of the improvements to the developer workflow could be open sourced beyond Sapling now that Phabricator is no longer supported.

Blender has been an amazing open source project - giving everyone a free alternative to create 3d content.

I also love how it’s relatively easy to script in python & you can even use a blender kernel to open in a Jupyter notebook for experimentation.

Prototyped in 10 days sounds more appropriate, then longer road to getting Javascript 1.0 publicly released.

What I'm glad about is that they kept improving the language, especially after ES6 where ECMAScript proposals can be independently evaluated & validated for backwards compatibility.

For a more UI / Web based slant, I would recommend these (beyond a Spreadsheet - which is incredibly helpful for understanding data flow systems):

* Simple video game using Unity or Unreal (understanding perf limitations in a game where 30-60 fps is critical helps make performant interfaces elsewhere - even on Web)

* A simple Javascript framework similar to React (again, will help understand data flow & handling events)

* An http library wrapper around XMLHTTPRequest (fetch exists now, but understanding how to send & read HTTP requests from scratch will help in debugging any problems down the line like CORS issues, OPTIONS requests, etc.).

Is there an archive of ipas that would work on the version of iOS that this is emulating?

I was just thinking about this because of the Nintendo server shutdown and it feels similar for iOS apps that only worked on previous versions or require specific servers to hit - wondering if a lot of these older apps & services will just be lost. Even services working today would have the same issue a decade from now if there’s not some standard to archive the data & api layers as well as the apps themselves.

One of the problems is that social media companies are incentivized to increase social media usage - whether watch time (YouTube, Facebook, TikTok), engagement (Twitter, Reddit, etc), or other metrics.

The government doesn’t have legislation on this and the Surgeon general warning is just advice at the moment - unless there’s a concerted effort like the Tobacco campaigns, this probably won’t get better.

This seems like it's only going to get worse many games need online services to function at all. Super Mario Maker as mentioned in the post is one of them, but there are countless others on 3DS and PC that won't function properly.

Ideally it would be possible to archive the games (like ROMs or what the Internet Archive does for websites and files) and also the server / database so that the games could be playable in the future, but I doubt that Nintendo or any larger publisher would do that.

I wonder if there's a solution for this in the short term or if these games and their data would be lost forever.

I think that it would also make sense to have a diagram that has open source achievements in Ai for 2023 (at least open weight & inference code). A lot of the announcements in the chart are behind an api so can’t be run locally.

Just thinking off the top of my head, Segment Anything, Llama 1 and 2, Mistral, Stable diffusion XL, ControlNet, Whisper are all open source AI releases this year.

Some results published here https://llm-eval.github.io/pages/leaderboard/advprompt.html

Also, technical report on arxiv https://arxiv.org/abs/2312.07910

I think that something that would make the results better would be highlighting which models perform best for specific tests (ie color coding) and explaining the tests via some hover info.

Also with some fine tuned models training to get higher scores on specific tests, I don't know how valuable these tests are in comparison to chatbot arena's elo ranking https://huggingface.co/spaces/lmsys/chatbot-arena-leaderboar...

So how would someone go through the process of invalidating patents like this? I assume that there's no easy or cheap way otherwise the eff would have already done it.

Another post mentions spending $500k for an IPR to invalidate - would that be the only way, limiting it to only well funded startups or larger established companies?

There's an online demo on huggingface here: https://huggingface.co/spaces/xichenhku/AnyDoor-online

You need to manually input segmentation at the moment, it would be helpful to use something like Segment Anything to automatically perform segmentation than having to do it manually.

Hopefully Windows 12's emulation of x86-64 apps is better than the support in the Windows 11 on ARM releases - if that's the case I could definitely see the next few years being a shift for Windows laptops from x86 to ARM with Nvidia also trying to make their own ARM chips.

I don't think it'll be as seamless as Apple's switch to Apple Silicon with Rosetta though.

This would make sense since the Gallup survey is based on people's perceptions based on social & traditional media that they see regularly. People remember notable events or shocking things, no one remembers the regular days where everything just went as normal - and the media doesn't cover such things since they are looking to grab attention with headlines.

The article mentions this at the end as outlier incidents, but just giving people the overall statistics may not help change their mind either because of what they perceive from other media sources.

I started getting into image generation last year with Stable Diffusion's 1.5 release last October, but seeing the improvements from Midjourney and DallE 3 is extremely impressive. The image to video models from Runway and Pika labs & Stable Video Diffusion are reminding me of Midjourney prior to V4, hoping to see similar improvements in 2024 for those models.

The issue with this is that we don’t know when the next recession is going to happen.

There could still be a recession in 2024 if the fed doesn’t get their “soft landing” while doing their planned 3 rate cuts.

I loved Lufia 2 as well, especially the Ancient Cave which blew my mind when I was younger since I didn't realize how they could make a full game and a replay-able 100 level dungeon in the same SNES cart.

My favorite game from SNES era is probably Tales of Phantasia though (the DeJap fan translation) - the gameplay mixed with a fun story across 2 different time periods was an amazing experience. I probably have FF5, 6, Chrono Trigger, Fire Emblem 4, or Star Ocean somewhere in the 2-6 range.

Local AI models (LLMs, Diffusion models) will take off in popularity as their performance will be good enough for many use cases without having to use a server. Also would allow a lot of startups to leverage client computation without having to host models on expensive GPU servers.

I'm eagerly waiting for 7 Remake's part 2 this February, having not played the original since it was on Playstation & I only had an N64. Having played a lot of SNES JRPGs via SNES9X or ZSNES, I loved the story and character building in FF6, gameplay of Tales of Phantasia and Star Ocean, and overall gameplay of Lufia 2 and Chrono Trigger when I was younger.

I think back and I find that not that many modern games have the same influence on me as those older games from my childhood - I don't think it's necessarily the quality of games has gone down, FF16, Tales of Berseria, YS VIII are all exceptional RPGs, but it might be that we have so many entertainment options these days that truly memorable games get lost when there's always something new around the corner.

Nice job, was going to mention url size limits but then read your blog post and it already mentioned using urls to be able to share with others without a database.

One thing that might be interesting is being able to edit things in the future if you want to plan something out, but don't know if that's really in the spirit of the site.

The numbers in the article are quite alarming for some countries - 49.9 out of 100 Canadians dependent (so one dependent for every 2 workers according to article) by 2075, or 77 out of 100 for Japan by 2075.

This seems like such a critical issue that governments around the world will need to have significant societal changes - at the moment this doesn't seem like something that tech or biotech will be able to solve simply.

Code is at https://github.com/Kevin-thu/DiffMorpher and github page is https://kevin-thu.github.io/DiffMorpher_page/. I think the github page shows more examples, so if it's possible to edit the link to that, it might be more helpful.