copypaper

Tron 1 for the plot, Tron 2 for daft punk, Tron 3... we don't talk about.

What a disaster and complete failure on the local government in the way they handled this situation. If we ever get hit by an earthquake or other larger disaster, it's safe to assume we're all on our own.

Also, as someone affected by this, it has been extremely frustrating getting updates via xitter. Do we really have no other options?

Yea after skimming the samples on Amazon I noticed that nearly every single sentence had at least one comma in it (adding zero value). It feels like I'm reading someones thoughts.

Personally, I love abusing commas for comments and shitposting, but they should be avoided in informative resources like books, otherwise, it looks like a word salad. Say your thoughts and ideas with boldness and certainty.

But hey you write better than I did at 18, so I ain't judging. Just trying to provide helpful feedback for you (the op) to improve on.

Same reason why WordPress is the de facto for businesses. You can create a "full" website with a few clicks and add some plugins to make it look complete despite it running like shit. It's all perception.

I just have a workstation at home that I SSH into from whatever device I feel like (within my tailnet). All my tools available on the CLI and vscode available via remote-ssh. You can connect from an iPad, macbook, chromebook, etc. The only thing it doesn't handle well is creative apps (video editing, blender, photoshop, etc.).

Obviously this whole setup requires an internet connection, but I'm rarely without one so it works great for me. Anyone else do something similar?

>25k commits in 4 months or about 1 commit every 7 minutes

How do you manage/orchestrate this? I'm genuinely curious.

touché hahah. Are there any SoTA open-source models that don't have corporate interest?

Pandora's box has already been opened and there is no going back. I doubt OpenAI, et al will get anything but a slap on the wrist in court because punishing AI companies would have a negative effect on the US economy.

>Can the same be said about DeepSeek or any other open-source model provider performing distillation?

Open source models that distill from SoTA reminds me of the story of Robin Hood -- robbing the rich and giving it to the poor. So to answer your question: yes, but it's better than the alternative where only a select few companies have SoTA models.

Let's also not forget SoTA models stole from us.

> we could super cool them and keep that at that temperature easily

As far as im aware, 3d stacking chips requires the inside to be cooled as well (not just the outside). I don't think they've solved this yet.

We are one technological breakthrough away from AGI. Seriously what happens when, for example, a viable room temperature superconductor (remember LK-99 lol) gets discovered? Next thing you know we have 3d stacked chips operating at THz speeds with virtually zero heat output, batteries that can charge instantly, etc.

I know a RTSC is the holy grail, but it really feels like AI is in the same stage computers were in the 80s. I used to be extremely bearish and think AI was useless, but I've taken a total 180 the last 6 months. If these things get better (they will), nobody's job will be safe.

In all seriousness, what is the game plan for society moving forward as AI takes more jobs? The government doesn't seem to care. The AI labs don't seem to care.

What happens when more and more people can't afford housing, kids, food, health insurance, etc.? Nothing more dangerous than a man who has no reason to live...

I don't advocate for violence, but I do foresee more headlines like this as things get worse.

This is awesome! I did a similar project in college for one of my classes and ran into the same exact walls as you.

- The more filters I added the worse it got. A simple EMA with smoothing gave the best results. Although, your pipeline looks way better than what I came up with!

- I ended up using the Teensy 4.0 which let me do real time FFT and post processing in less than 10ms (I want to say it was ~1ms but I can't recall; it's been a while). If anyone goes down this path I'd heavily recommend checking out the teensy. It removes the need for a raspi or computer. Plus, Paul is an absolute genius and his work is beyond amazing [1].

- I started out with non-addressable LEDs also. I attempted to switch to WS2812's as well, but couldn't find a decent algorithm to make it look good. Yours came out really well! Kudos.

- Putting the leds inside of an LED strip diffuser channel made the biggest difference. I spent so long trying to smooth it out getting it to look good when a simple diffuser was all I needed (I love the paper diffuser you made).

RE: What's Still Missing: I came to a similar conclusion as well. Manually programmed animation sequences are unparalleled. I worked as a stagehand in college and saw what went into their shows. It was insane. I think the only way to have that same WOW factor is via pre-processing. I worked on this before AI was feasible, but if I were to take another stab at it I would attempt to do it with something like TinyML. I don't think real time is possible with this approach. Although, maybe you could buffer the audio with a slight delay? I know what I'll be doing this weekend... lol.

Again, great work. To those who also go down this rabbit hole: good luck.

[1]: https://www.pjrc.com/

Yea, but can it secure systems from the unpatchable $5 wrench vulnerability?

https://xkcd.com/538/

Mainly 09, but also 05 and 07.

Very interesting. Im on vacation but will check this out at work next week.

What is the maximum resolution you support for PDFs? The max gemini will do is 3072x3072. We have plans that are 10x that size.

Indeed it was (I was listening to it while stumbling across this post). Also, fun fact: The Gambler was written by Don Schlitz while working as a Computer Operator in 76' which makes it all the more relevant [1].

[1]: https://web.archive.org/web/20230130060050/https://www.rolli...

You got to know when to Ship it,

Know when to Re-prompt,

Know when to Clear the Context,

And know when to RLHF.

You never trust the Output,

When you’re staring at the diff view,

There’ll (not) be time enough for Fixing,

When the Tokens are all spent.

We have the technology, its just heavily despised due to the lack of privacy and anonymity.

Gemini on the web as a chat app is great (as well as NotebookLM). But Antigravity is an embarrassment and the reason I cancelled my Gemini subscription. I'd recommend avoiding it at all costs. It has degraded so badly in the last two weeks it's hardly usable anymore.

The difference here though is that ads are baked into the response via plain text.

How far away are we from an offline model based ad blocker? Imagine a model trained to detect if a response contains ads or not and blocked it on the fly. Im not sure how else you could block ads embedded into responses.

Nice work! I'm curious though, what was your use case for needing a smaller library? Since you're running this on a server, what difference does an extra 226KB make?

The article also lacks any personal opinion or experience on the matter. It just stated a bunch of pointless stats and "facts". Almost every single point is refutable depending on the task at hand.

Yep. My grandma bought her house in ~1962 for $20k working at a factory making $2/hr. Her mortgage was $100/m; about 1 weeks worth of pay. $2/hr then is the equivalent of ~$21/hr today.

If you were to buy that same house today, your mortgage would be about $5100/m-- about 6 weeks of pay.

And the reason is exactly what you're saying: the average US worker doesn't provide as much value anymore. Just as her factory job got optimized/automated, AI is going to do the same for many. Tech workers were expensive for a while and now they're not. The problem is that there seems to be less and less opportunity where one can bring value. The only true winners are the factory owners and AI providers in this scenario. The only chance anybody has right now is to cut the middleman out, start their own business, and pray it takes off.

>Each model gets access to market data, news APIs, company financials...

The article is very very vague on their methodology (unless I missed it somewhere else?). All I read was, "we gave AI access to market data and forced it to make trades". How often did these models run? Once a day? In a loop continuously? Did it have access to indicators (such as RSI)? Could it do arbitrary calculations with raw data? Etc...

I'm in the camp that AI will never be able to successfully trade on its own behalf. I know a couple of successful traders (and many unsuccessful!), and it took them years of learning and understanding before breaking even. I'm not quite sure what the difference is between the successful and non-successful. Some sort of subconscious knowledge from staring at charts all day? A level of intuition? Regardless, it's more than just market data and news.

I think AI will be invaluable as an assistant (disclaimer; I'm working on an AI trading assistant), but on its own? Never. Some things simply simply can't be solved with AI and I think this is one of them. I'm open to being wrong, but nothing has convinced me otherwise.

Every single SDK I've used was a nightmare once you get past the basics. I ended up just using an OpenRouter client library [1] and writing agents by hand without an abstraction layer. Is it a little more boilerplatey? Yea. Does it take more LoC to write? Yea. Is it worth it? 100%. Despite writing more code, the mental model is much easier (personally) to follow and understand.

As for the actual agent I just do the following:

- Get metadata from initial query

- Pass relevant metadata to agent

- Agent is a reasoning model with tools and output

- Agent runs in a loop (max of n times). It will reason which tool calls to use

- If there is a tool call, execute it and continue the loop

- Once the agent outputs content, the loop is effectively finished and you have your output

This is effectively a ReAct agent. Thanks to the reasoning being built in, you don't need an additional evaluator step.

Tools can be anything. It can be a subagent with subagents, a database query, etc. Need to do an agent handoff? Just output the result of the agent into a different agent. You don't need an sdk to do a workflow.

I've tried some other SDKs/frameworks (Eino and langchaingo), and personally found it quicker to do it manually (as described above) than fight against the framework.

[1]: https://github.com/reVrost/go-openrouter

Regardless of which approach ends up being right, the tool itself is amazing. Best of luck with it!

This looks awesome!

>I needed it to be a pretty fully functional IDE. I needed [IDE feature]...

I'm just curious, why didn't you make this as a VS Code plugin to benefit from all the features of an IDE? I'd imagine you could do something similar to the Live Server plugin. That way you could support any browser and not worry about maintaining the IDE features.

Wow I didn't know he got laid off. I remember watching his videos a few years ago and absolutely loved his way of teaching. He's the reason I found out about PlanetScale to begin with.

Looking at their youtube channel, Aaron's videos had a total of ~1.4m views over 24 videos (an average of ~58k per video). Their recent videos don't even get past 1k views...

Yea exactly. The LLMs are tuned to natural language. I don't think anything will beat good ol' templating (a.k.a. plain text). In Go I do something like this:

  // mytemplate.tmpl
  Description="The following data is for the users in our application."
  Format="id,name,role"
  length=2
  Data:
  {{range .}}
  {{.ID}}, {{.Name}}, {{.Role}}
  {{end}}

This way you're able to change the formatting to something the LLM understands for each struct. The LLM might understand some structs better as JSON, others as YAML, and others in an arbitrary format. Templating gives you the most flexibility to choose which one will work best.