zlurker

JEV directly addresses many of the most common issues with LLMs for certain applications. It seems IMO to be overhyped right now but I think it may, like the broader llm ecosystem, be here to stay.

They spent months hyping up Mythos and ended up with it banned. I’d assume they want to both differentiate their products and appeal to regulators here

I still dont fully understand the point of MCP servers. What do they provide that a skill doesnt? Maybe I've just used too many poorly written ones.

Is there some sort of tool that can be expressed as an MCP and but not as an API or CLI command? Obviously we shouldnt map existing apis to MCP tools, but why would I used an MCP over just writing a new "agentic ready" api route?

How is this different than adding a calendar mcp server to something like Claude?

Timezones still play a huge part here as well. My team has several engineers who are at par compared to the rest of the team but they're a pain to work with due to a 10 Hour time difference.They may be paid 50% less but they're 50% harder to work with to no real fault of their own

interesting! what sort of errors did you find and were there any sort of take aways your had other than lack of care by the data provider?

even with the supported plug-ins it's still not the same either. intellij with python plugin has just enough differences with it that it throws off some of my coworkers who use pycharm.

Yep, I got intellij to work with python/rust/scala and it's frustrating to see it pulled out like that. Not sure why they need to sell so many IDEs that are almost identical. Would've been happy with some upcharge for a "premium" plugin but I don't want to install three IDEs like they apparently want me to.

Errors/Exceptions by value isn't that important in isolation, but it makes functional programming styles work a lot better (monads).

I think there's still an argument here that this is just a symptom of trying to force functional paradigms on python, which is happening more and more, which I like but am sometimes worried about how pythonic it is... For a one odd script it works well, but when I want to maintain something and have to manage many dependencies and onboard new devs it becomes quite challenging.

I think this over complication makes the comparison more realistic. Yes for this exact example it's too much, but it gives a better understanding of what a more complex real-world example would look like.

Yeah, dagster has a concept of metadata and assets so we have some code that'll map dagster's own logical representation to physical s3 locations.

Reports and viz varies a lot, the finance department uses tableau where as for more 'data sciencey' stuff we normally just use notebooks.

We orchestrate our ETL pipelines with dagster. We only use duckdb in a few of them but are slowly replacing pandas etls with it. For some of our bigger jobs we use spark instead.

Essentially it's: 1. Data sources from places such as s3, sftp, rds 2. Use duckdb to load most of these with only extensions (I dont believe there's one for sftp, so we just have some python code to pull the files out.) 3. transform the data however we'd like with duckdb. 4. convert the duckdb table to pyarrow 5. Save to s3 with delta-rs

FWIW, we also have this all execute externally from our orchestration on an EC2 instance. This allows us to scale vertically.

It's a bit frustrating reading this article, There's clearly two different problems going on here.

Claiming your filling is 100% tuna or that your drink is made with aged vanilla, when both are completely untrue are just outright lies to deceive people. Companies doing this should legally be required to change their ways.

On the other hand, redbull not literally giving you wings is obvious. Only a child would literally believe this. There's a stark contrast between this and expecting açai in a "strawberry açai" drink.