aaronscott

I think their solution is still sub-optimal. Ideally a secondary agent would pre-process the prompt, select relevant information from the catalogue, then pass that on to the primary agent.

This way the primary agent only has relevant information in their context to make decisions and take actions.

Context management is still under valued imo.

Something that others haven’t mentioned is that you can flash an open source OS/reader software (Crosspoint, among others) so you can change the UI and experience for your preferences. I like having an e-reader that lets me modify how it works.

I replaced the battery on my kindle 4 not that long ago. It had the best UX for extended reading when compared to newer kindles.

The biggest downside was not having a frontlit display.

I recently switch to an xteink x4, and found that several others in that community migrated from kindle 4s as well. So there are still some number of users in the world that value the device.

Same here, I quite enjoy it. Plus there is open source software available, such as crosspoint. It’s easy to flash and an opus call away to change the behavior if you want something to work differently.

This is awesome!

Thanks for explaining the reasoning behind implementing your own version instead of using Al Dente. The straightforward UI is really appealing.

I paid for an Al Dente license long ago and have been very happy with it, sail mode and temperature sensitive charging are great features (although I’m not sure how big of an impact sail mode makes).

It’s great to see alternatives available though. Having command line access is really cool.

They have an ultrawide mode available now. Personally I find it very uncomfortable. You have to move your whole head to see the sides, and the vision pro is heavy. Looking off to the side for a length of time is uncomfortable.

It would be nice if the search provider could be configured. I would like to use this with Kagi.

They have a free demo that is fairly extensive. It should give you a good feel for whether the game is worth it to you.

If you find this topic interesting, David Imel put together a fantastic deep dive. He spent four months of research, went to Japan to interview the creator of emoji, and put together a very good video on the topic[0].

[0] https://youtu.be/g-pG79LOtMw?si=htnop9jpjmE3kQ75

> I like to define my subroutines using a modern language like C++, which goes 47 gigaflops. This means C++ is three orders of a magnitude faster than Python. That's twenty years of progress per Moore's law.

This is great. I love the idea of measuring performance differences in “years of Moore’s law.”

Twenty years puts the delta in an easy to understand framework.

I agree with your observation, although it seems unpopular at the moment. It’s clear the video is slowed down and seeing how his hands respond at the end of his action, it looks like he’s throwing the wheel very quickly.

The tires start moving immediately, so it looks more like a limit in how quickly they change direction. It appears laggy because they don’t finish the movement when he does. They do respond right away when he shifts his arms a little in the middle of the video.

Like others, wanted to say thank you for writing Aider.

I think you've done a fantastic job of covering chat and confirmation use cases with the current features. Comments on here may not reflect the high satisfaction levels of most of your software users :)

Aider helps put into practice the use cases that antirez refers to in their article. Especially as someone get's better at "asking LLMs the right questions" as antirez refers to it.

This is amazing, thanks for sharing it! I’ve been thinking about building something along these lines, so it’s great to see a working model.

I completely agree. A huge business with a singular focus isn’t going to pivot into the music business (or any of the myriad use cases the general public throws at it). And if they did use someone’s info, it’s more likely an unethical employee than a genuine business tactic.

Besides, the parent program uses the API, which allows opting out of training or retaining that data.

During a job interview I was asked to teach something off the cuff. There were two interviewers in the meeting, so I decided to show the difference that a good mic with good placement can have on call quality. Switched from my nice mic that was one shaka[0] away, to the laptop mic that was about two arm lengths away (maybe 8 shaka’s). They were shocked at the difference and one asked “is that what I sound like right now?”

I went on to discuss how sound intensity is inverse to the square of the distance, so double the mic distance and it’s 1/4 the intensity. So the distant mic was picking up ~1/64th the volume. When there isn’t much of a difference between my voice and room echoes it becomes hard to listen to.

They started a discussion about how much it would cost to outfit everyone with external mics after that haha. For anyone wondering, even a cheap lapel or broadcast mic that’s close to your face will sound much better than a distant laptop, due to the comparative volume of your voice verses the environment.

[0] One shaka, or one “hell yeah” is a unit of measure an audio engineer shared with me once. Extend your thumb and pinky, and get your mic about that far from your mouth for good sound. One or two shaka’s is generally very good.

I would be very interested, depending on cost.

For pricing models, JetBrains (PyCharm) has a nice setup, although it could be more user friendly. After 12 months of payments you get a perpetual fallback license[0].

The more user friendly version would be getting perpetual access to the current version of software after 12 months of payments, rather than the backdated version. But I understand their reasoning.

0 - https://sales.jetbrains.com/hc/en-gb/articles/207240845-What...

Also a very satisfied customer (since 2015). I’ve used this to do everything from making clunky software easier to use to automating aspects of games that I didn’t want to do manually.

Clunky software example: connect to cisco vpn each day, handling all the prompts it makes you go through.

Another is to type out what’s in my copy buffer, for websites that don’t let you paste your password in.

Gaming example was re-casting spells when they came off cooldown in Diablo 2/3.

I’ve written hundreds of macros over the years and this app has saved me a ton of time and frustration. And probably some RSI too.

Your comment about being continuously narked reminded me how in early human history people relied on low-alcohol beer as a safe water supply. So back then it was a double-whammy.

https://www.scientificamerican.com/article/the-conflicted-hi...

I was curious for a summary of what the book suggested. Here is Kagi’s summarizer:

- Emotional maturity and stability are key traits for a happy marriage. People who are well-adjusted, have good self-confidence, and can handle problems well tend to have more successful marriages.

- Communication and companionship are vital for a happy marriage. Couples need to develop a strong bond through sharing interests, aspirations, and conversations.

- Compatibility and similar traits between partners promote marital happiness. Couples who have similar personalities, backgrounds, and interests tend to get along better.

- Sexual satisfaction and intimacy are important for a fulfilling marriage. Couples need to develop a good sexual relationship and intimacy to avoid infidelity and build closeness.

- Mutual understanding, trust, and respect are crucial for a lasting marriage. Partners need to be considerate of each other's needs, supportive, and willing to compromise.

- Shared goals and values help strengthen a marriage. Couples who have common life goals and moral values tend to be more content in their relationships.

- Financial stability and compatibility aid marital happiness. Couples who manage their finances well and agree on spending habits tend to have fewer conflicts.

- Physical and mental health issues can strain a marriage. Conditions like addiction, mental illness, and disabilities require effort and understanding from both partners.

- Similar family backgrounds promote marital success. Children of happily married parents tend to have more successful marriages themselves.

- Premarital counseling can help resolve issues before marriage. Seeking advice from professionals and discussing expectations can prepare couples for the challenges of married life.

> Another simple risk mitigation step we take, that we believe to be unique to OceanGate is that we draw a small vacuum on the inside of the sub at the start of each dive. This step verifies the integrity of the low-pressure O-ring seal and eliminates the risk of leaks

I wonder what that low-pressure o-ring is sealing. I assume the vacuum would only simulate a one atmosphere differential, so that o-ring must not be sealing something exposed to the external pressures at the depths they go down to.

You can download a model from a site like huggingface. Here is a list of models that can be used for inference:

https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderb...

This user has some models already compiled for use with GGML (look for models with that in the name):

https://huggingface.co/TheBloke

Or if you want to convert your own model the llama.cpp repo has good instructions. Briefly it's `python3 convert.py <model>` - then if you are using a large parameter model you may need to quantize it to fit in memory `./quantize <source_model> <destination_name> <quantization>`

I think the exciting thing is that it works as GPU memory. 192gb would be great for running LLMs.

https://news.ycombinator.com/item?id=36187466

For applied maths I agree with you. But when exploring novel and innovative approaches I disagree. There is so much we can learn from finding new ways to look at the world. The initial take may be complex or clever, but it can lead to a deeper understanding that allows for later simplification of entire subject areas.

I have wondered about sensory input being needed for AGI when thinking about human development and feral children[1]. It seems that complex sensory input, like speech, may be a component of cognitive development.

https://en.wikipedia.org/wiki/Feral_child

It’s been really helpful for picking up a new programming language. Particularly around helping me understand conventions in a language I’m not familiar with.

GPT-4 has been great at breaking down complex regexes that I am too lazy to parse out in the moment.

I’m also finding it helpful as a creative partner on naming things. Something I feel like I spend a lot of time noodling on. Like: creative names for a data warehouse that are surfing related (BoardRoomData lol).

I want to reinforce your plan. Somehow I played factorio for a few hundred hours before finding online communities and blueprints. It was a wonderful experience of discovery and exploration. I came up with a few "novel" designs during that time that I'm still proud of.

Once comparing my own work to the best online it took a lot of joy out of the game. But I found that the memory was short lived, once I cut myself off from those online communities it was easy to fall back into the "let me pull out my sketchbook and solve this problem" mindset. I think half the fun is that design process.

I'm well north of 1000 hours now, and still discovering new ways to play and solve problems. I've come full circle and now build spaghetti bases because they are so visually fun.

Your reference to gas powered cars going that way (a bit more slowly) made me wonder about how quickly some product lines were disrupted in different industries.

First mass produced digital SLR was by Nikon in 1988[0]. So this would represent roughly 34 years to fully replace the product line with digital cameras.

First mass produced electric car would be either Tesla Model S (2012)[1] or their Roadster (2009)[2]. I'm not sure if 147 Roadster's count as mass production. So if gas cars are being fully dropped by 2043 or 2046 then it'd be on the same scale as the digital camera.

Feature phones being dropped in favor of smart phones was pretty quick. I'm not sure when major manufacturers stopped feature phone production, Nokia 3310 had a revival in 2017. So more than 10 years, but probably less than 15.

0 - https://en.wikipedia.org/wiki/Digital_camera#cite_note-busch...

1 - https://en.wikipedia.org/wiki/Tesla,_Inc.#cite_note-40

2 - https://en.wikipedia.org/wiki/Tesla,_Inc.#cite_note-32

I looked into buying one a few weeks ago but ultimately decided not to move forward. One reason was that it comes with a pre-configured list of 500 chords, so it's up to users to create their own beyond the initial list. The other was that the build quality doesn't appear to be that great (with the PCB board clearly visible).

The reddit community[0] has some helpful reviews and progress updates. And their discord server[1] is quite active with lots of helpful links and advice.

0 - https://www.reddit.com/r/charachorder/ 1 - https://discord.com/invite/jMj6grUuBc

Just seconding this. They have had stock fairly regularly, and they have a pretty good notification email for when they will have new items available. I purchased a 3090 from them, along with a few other parts.

Regarding buying the mod that goes into space, they did one better and hired the developer of that mod.

From the blog post:

> One of the steps we have already taken is hiring a concept artist, he might be known to you under the nick of Earendel

Earendel's list of mods for those curious: https://mods.factorio.com/user/Earendel