totalhack

What is the latency you are seeing with 5.4 no reasoning? And where have you landed for stt and tts solutions?

The tiring part for me is the waiting and/or context switching to fill that time. When agents of this intelligence or better can get you results in seconds instead of minutes we can start thinking single threaded again and it will be more enjoyable.

If you are talking about the Copilot built into vs code, that's not been my recent experience at all. Very capable in agent mode since gpt 5.4 came out.

It's optional, and that's great, because its necessity or benefit is situational.

Agree with this, though I've mostly been using Gemini CLI. Some of the simplest things, like applying a small diff, take many minutes as it loses track of the current file state and takes minutes to figure it out or fail entirely.

I use markdown files in vs code. Supports folding of sections, combined with a few keyword/char patterns that I highlight different colors. Also get code highlighting in properly labeled code sections.

Did you find out what service he was using?

Well, pretty much everyone is less valuable than they think!

I've bootstrapped and sold two startups, first time with 5 (!) equal cofounders and second time with 3. I've learned a lot along the way, obviously some of that through mistakes of my own. Have my cofounders been perfect? Heck no -- one lacks the same level of drive, the other is less trustworthy (always multiple irons in the fire kinda guy). But we all have strengths, are all generally better off together, and have the combined skill set to run super lean. Even if you think you can do some things better alone, you can never do it all, and that grind will take quite a toll.

Related: over the years the topic of going into business with some of my best friends has come up, and I always say I would never do that. Running a fast-moving, high stakes business with multiple founders involves a lot of tough conversations and occasional relationship strain. My business partners are business partners first, friends second, and I think that's important to consider if you really value a friendship.

I just went through this process and ended up getting the HP Omnibook Ultra Flip with a core 7 258v, running Ubuntu 24.10. Performance is excellent, battery life is the best I've ever had on an Intel running Linux. If you truly need more RAM than 32 GB it won't be an option though.

The air or an mb pro seems so nice until I remember the sting of dealing with the dev environment and docker on a Mac compared to Linux. No amount of battery life or marginal jump in performance (which gets lost through needed virtualization) will make up for that for me.

There are a number of semantic layer tools out there these days. Each has their own unique approach, but essentially it's a meta layer on top of your database that can be used to do things like form queries or provide a consolidated API to your data (which may be in multiple databases).

Some comments on this thread mention popular semantic layer tools like cube.dev. I also made an open source one that I use regularly, though it's currently in I-hope-to-put-more-time-into-this-someday mode. Been busy with an acquisition this year.

https://github.com/totalhack/zillion

You are doing a lot of the work a semantic layer would do for you. I wonder if you would have better luck having the LLM talk to a semantic layer instead of directly to the database.

This is slightly less interesting now that I can have AI remember bash script syntax for me.

Happy to see them announce the new AMD chips in the 13", but the prices are a little nuts compared to what you can get elsewhere.

I bet the desktop is interesting, but first I was in a 30m waiting room to access their website (what?!?) and then the button to pre order the desktop is broken. Ouch.

The constant struggle is the lagging support for the best hardware. Unless you want to buy a machine from a dedicated Linux shop (framework, System76, tuxedo), you are taking a risk buying anything remotely new without a lot of research into the hardware and hunting for posts about people who have had success. Those dedicated Linux shops are usually at least a generation behind too. It doesn't help that architectures are evolving rapidly in response to Apple's chips (not a bad thing long term).

Just bought an HP omnibook flip with an Intel 258v. From what I've read I expect most things to work, or work soon with kernel updates. All applications I care about work well. Would be nice to see OpenAI make a desktop app for Linux, or some other open source equivalent.

Thanks for taking a look! Happy to chat, DM me here: https://bsky.app/profile/totalhack.bsky.social

I have nothing against SQL of course. The simplified approach of a UI built on top of zillion or tools like it really enables a whole next level of productivity for business users that are never going to learn SQL, but also need more query flexibility than just "dashboards" without having to wait on a BI team for answers -- I die inside a little bit every time I hear of a company doing this. And as you have noted, I also think text-to-semantic-layer is an interesting approach for involving AI/NLP.

I've been pulled away from this project for some time due to an acquisition at my day job but hoping to get back into it soon!

Congrats on the launch. I made a tool that has some similar objectives but doesn't present as SQL itself like Trilogy seems to. I'll take a deeper look at Trilogy soon, always interested to see the variety of approaches to this.

https://github.com/totalhack/zillion

If I'm understanding your intent correctly, I think you could do that with Zillion.

https://github.com/totalhack/zillion

Disclaimer: been sidetracked by an acquisition at my day job this year, intend to put more time into this project soon, but I use it in production to great effect.

Running a 2024 xps 13 with Ubuntu for work and it's been solid. Had a Lenovo before this which was great bang for the buck but occasional issues with heating up during sleep. Would consider trying a Framework next.

Expect was the first tool that really hooked me at my first job at Sun. I had to write a tool to automate verification of the boot process during firmware dev, handling variety of odd/unexpected behaviors, paths, and prompts.

Many years later I heard they were still relying on that tool!

Is this more like text-to-semantic layer or does it throw the schema in the prompt and generate SQL with the llm?

Go f'in get 'em Scarlett.

This looks cool. I built a similar open source semantic data / warehousing tool called Zillion. I use it to power my company's BI but haven't put as much time into the polish as you guys.

https://github.com/totalhack/zillion

I'm not opposed to any attempt to remove the need for these full college degrees as a hiring requirement, but this also ignores that the point of many colleges is the network. You are paying to be part of a club.

Loans. Loans are how this happened.

I also assumed this was an April fool's joke from the other terraform.

Zillion queries happen in two layers...

It first runs one or more queries against your DataSources in a drill-across query fashion. You can think of DataSources as one or more completely separate databases. You could have one mysql, one postgresql, one duckdb etc all in the same Warehouse (not saying this is common in production, just an example). Within those DataSource queries it's also joining all needed tables together for you, i.e. joining multiple tables in each database to meet your required grain.

It then takes the results of all those queries and combines that data in another layer which is currently an in-memory sqlite database. The purpose of that layer is joining the data for presentation as well as applying some additional features like rollups, technicals, formula fields, etc.

I'm not familiar with what superset does under the hood or exposes as an API so I don't know how to compare it, if there is some similar backend piece. But I suspect no part of superset is quite the same as this, based on what its front end can do.

Happy to answer any other questions you have.

The lack of a semantic layer and join limitations are what made me pass on superset, but that was a couple years ago so maybe those features have been added.

I built my own semantic layer instead. I use this in production in my company but obviously use at your own risk as it's a one-man show.

https://github.com/totalhack/zillion

I thought this was some jargon I didn't know haha.

Superset would be on my shortlist if I had to use something else, but the join limitations were part of why I passed.

Thanks! I did notice the queue count showing up occasionally but not every time. Maybe someone could repeat the test who has access without the queue so we can get an understanding of the potential latency once scaled and geo-distributed. What I'm really trying to understand is time to first token output actually faster than GPT 3.5 via API or just the rate of token output once it begins.