planckscnst

I think this is a valuable direction for research to take. I suspect human emotion is represented in the training data and that there is some underlying model representing human emotional systems present within current LLMs. If we can understand how the model affects response given certain qualities in provided stimuli, then we can effectively use this embedded model to improve our understanding of both LLMs and Humans and also get more productive output from LLMs.

I've found this orchestrator+reviewer+judge setup to yield much better results than anything else I've tried. And it's such a simple setup - a few markdown files.

I'm also creating one that is similar, but purpose-built for making the plans that this setup can orchestrate. It still needs some tweaking to get agents to follow it better - it still takes additional prompting to nudge it down the proper path. But I've had similar benefits - sending plans through this adversarial review loop has yielded significant improvements in final output.

https://github.com/Vibecodelicious/llm-conductor/blob/main/p...

That's so strange. I found GPT to be abysmal at following instructions to the point of unusability for any direction-heavy role. I have a common workflow that involves an orchestrator that pretty much does nothing but follow some simple directions [1]. GPT flat-out cannot do this most basic task.

[1]: https://github.com/Vibecodelicious/llm-conductor/blob/main/O...

The API key is not a subscription. The title says subscriptions are blocked from using third-party tools. Or am I misunderstanding?

I've been (adding an OpenCode feature that allows the LLM to edit its own context)[1] and trying to debug an issue with the Anthropic API because I'm calling it with missing fields that it expects. I hope my multiple erronious API calls aren't what triggered this decision

[1]: https://github.com/Vibecodelicious/opencode/tree/surgical_co...

For now, this is mitigated by only including trusted content in the context; for instance, absolutely do not allow it to access general web content.

I suspect that as it becomes more economical to play with training your own models, people will get better at including obscured malicious content in data that will be used during training, which could cause the LLM to intrinsically carry a trigger/path that would cause malicious content to be output by the LLM under certain conditions.

And of course we have to worry about malicious content being added to sources that we trust, but that already exists - we as an industry typically pull in public repositories without a complete review of what we're pulling. We outsource the verification to the owners of the repository. Just as we currently have cases of malicious code sneaking into common libraries, we'll have malicious content targeted at LLMs

I've directly faced this problem with automatic code formatters, but it was back around Claude 3.5 and 3.7. It would consistently write nonconforming code - regardless of having context demanding proper formatting. This caused both extra turns/invocations with the LLM and would cause context issues - both filling the context with multiple variants of the file and also having a confounding/polluting/poisoning effect due to having these multiple variations.

I haven't had this problem in a while, but I expect current LLMs would probably handle those formatting instructions more closely than the 3.5 era.

LLMs are very good at looking at a change set and finding untested paths. As a standard part of my workflow, I always pass the LLM's work through a "reviewer", which is a fresh LLM session with instructions to review the uncommitted changes. I include instructions for reviewing test coverage.

I've also found that LLMs typically just partially implement a given task/story/spec/whatever. The reviewer stage will also notice a mismatch between the spec and the implementation.

I have an orchestrator bounce the flow back and forth between developing and reviewing until the review comes back clean, and only then do I bother to review its work. It saves so much time and frustration.

For interactive programs like this, I use tmux and mention "send-keys" and "capture-pane" and it's able to use it to drive an interactive program. My demo/poc for this is making the agent play 20 questions with another agent via tmux

I don't have a link to share just yet, but I'm working on an LLM coding agent that can modify its own context and is given hints on when and why that would be useful.

I expect it to make it possible to not think about when to reset back to a clean session. I also expect it to be more efficient as it will clear out all the "garbage context" that only serves to "confuse" the LLM, cost more tokens, make responses slower, etc.

Once I get a working prototype, then I will test the feature by using it while reimplementing it in other open source agents to get a feel for whether it has the effects I'm expecting.

If they were using this compression for storage on the cache layer, it could allow more videos closer to where they serve them, but they decide the. Back to webm or whatever before sending them to the client.

I don't think that's actually what's up, but I don't think it's completely ruled out either.

If you like this tool, you might also be interested in reptyr, which lets you reparent a process to a different tty.

https://blog.nelhage.com/2011/02/changing-ctty/

Whenever I use LLM-generated content, I get another LLM and pre-bias it by asking if it's familiar with common complaints about LLM generated content. And then I ask it to review the content and ask for it to identify those patterns in the content and rewrite it to avoid those. And only after that do I bother to give it a first read. That clearly didn't happen here. Current LLM models can produce much better content than this if you do that.

Definitely impossible in 2015.

If you're talking about this part:

> Initially, we were unable to add capacity to the metadata service because it was under such high load, preventing us from successfully making the requisite administrative requests.

It isn't about spinning up ec2 instances or provisioning hardware. It is about logically adding the capacity to the system. The metadata service is a storage service, so adding capacity necessitates data movement. There are a lot of things that need to happen to add capacity while maintaining data correctness and availability (mind at this point, it was still trying to fulfill all requests)

There's also dynamodb-fips.us-east-1.amazonaws.com if the main endpoint is having trouble. I'm not sure if this record was affected the same way during this event.

I don't remember an event like that, but I'm rather certain the scenario you described couldn't have happened in 2017.

The very large 2017 AWS outage originated in s3. Maybe you're thinking about a different event?

https://share.google/HBaV4ZMpxPEpnDvU9

I agree that financial viability is critical to the long-term prospects of a technology. It must deliver an ROI above other options. I'd recommend getting off the sidelines and jumping in to see what's happening. At the least, you'll have another perspective to inform your position. It's a pretty minimal investment to try it out.

I learned several languages before Python, but the one that made it the most difficult was Ruby. After having done Ruby for years and becoming really familiar with idioms and standard practices, coming to python feels like trying to cuddle with a porcupine.

List (or set, or dict) comprehensions are really cool... one level deep. The moment you do 'in...in' my brain shuts off. It takes me something like 30 minutes to finally get it, but then ten seconds later, I've lost it again. And there are a bunch of other things that just feel really awkward and uncomfortable in Python

Buildroot is easy to use even without vendor support

Yes, it feels wasteful compared to my previous driving, in specific circumstances. I drove a manual transmission and made it a goal to not touch the brakes at all for red lights (given acceptable traffic conditions - sometimes people would just be annoyed or confused if I did it every time). I'd see the light from far back, understand its timing, and how many cars were stopped behind it and gauge when I should disengage the clutch and coast. It frequently worked out that I was arriving at the stopped car in front of me just as they were starting to move forward. It feels particularly satisfying not to lose all your momentum just to have to accelerate again from a dead stop.

Overall, however, regenerative braking probably is better given the conditions don't always allow for this game.

You can also zoom in

I wonder if we can come up with an agreed name for almost json stuff... Maybe json+ and we can tac-on additional characters for what's added. json+c (plus comments), json+t (plus trailing commas), json+tc or json +ct (plus comments and trailing commas), and so on

> Society doesn't exist anywhere besides in your head.

How do you differentiate between what exists only in your head vs what exists elsewhere?

Societies make choices that have physical and pragmatic realities associated with them. Those choices impact our everyday lives.

You can also use some other part of your body that has moisture. My nose and my scalp are really oily, so I can rub my finger on my nose and then do the touch controls and fingerprint sensors and have it work

Computers aren't evolving in an energy-scarce environment (yet) we (currently) pretty much give computers as much energy as they want. Computers are evolving in an environment where a dependent species relies on them for that species' own weaknesses.

That one person who laughs so loud it's painful even though I'm 15 feet away, and they laugh like 3x per minute continuously... drives me up the wall. Yeah, I'd definitely appreciate a way to tune them out!

> we can achieve comparable performance to an in-memory database system for the cached working set

Should I keep reading, or is the title misleading?

The abstract seems to say that the system provides memory-comparable performance for data that is... in-memory

This is really cool! For those who take it as a joke or who scoff at the choices (like the whitespace thing), at least read until section 4 and then skip to section 9.

If you have a mid- to upper- range phone made in the last 4 years, that's probably the best low-budget option by far. Save your money and get led panel lights instead.

And a good audio recorder will do wonders. Zoom h1n is a quite good budget option. Try searching YouTube for "h1n vs" - there are tons of great options.

And it would be magic for storyboarding. This would be such a useful tool for a director to iterate on a shot and then communicate that to the team