Hari

Tech, football, misc in that order. Forgive typos and imperfect sentences, for I am no llm

[Hari @randome_dev]

Wrote and posted my first tech blog in medium pls do give it a read and suggest any improvements that should be done.

https://t.co/do08btIwfO

[Hari @randome_dev]

@alxshp Definitely going to try this, psql has many limitations and wasn't able to find a gui client to my liking

[Hari @randome_dev]

@PatCastiglia @jankeesvw Does this need hardware support too or is tgis a software level thing?

[Hari @randome_dev]

S3 must be in top 5 most influential thing ever built in tech.

[Neon @neondatabase]

Postgres on S3 is the right design, but even more interesting is making WAL the source of truth. This is how our storage works - we wrote a deep dive, summary on thread: https://t.co/7iHKt2RpNf

[Emil Kowalski @emilkowalski]

The thing I miss the most about coding by hand is the clarity.

You knew exactly what lives in which file, why you wrote a specific line, why you rewrote the other one, why you used this library.

You still can have such awareness, but it’s harder since everything happens faster

[Hari @randome_dev]

Well just inform us when it is up

[GitHub Status @githubstatus]

We are investigating reports of degraded performance for Actions https://t.co/WPsQRhAW0q

[Hari @randome_dev]

@AdamHoltererer They were already really good with images even before llms with Google lens. Also guess which org has the largest amount of image data.

[Hari @randome_dev]

@RhysSullivan Waiting for mcdonald code

[Hari @randome_dev]

When I am working on a task I try to mostly divide it into multiple phases of implementation or if its small single phase. Each phase implemented by a low reasoning cost effective model and final testing done by another. The main agent's job is to do the /grill-me part and creating spec file and that only should be a really good model.

This has been working for me well, very few auto compaction I have hit.

[Hari @randome_dev]

@Ananth7e I burned min in 3 hours

[Hari @randome_dev]

This is why we can't have good things. I think they already have removed this as last day my task did stop if I remember correctly. What a shame

[Tibo @thsottiaux]

@maria_rcks Hmmm. That’s unfortunate. We have this feature because it’s a nicer user experience overall.

Of course you can abuse it and trick the system somewhat, but if done at scale then it will force us to revisit this nice feature.

[Hari @randome_dev]

Whatever, man. I wouldn’t touch Opus 5 or Sonnet 5 with a ten-foot pole. In claude code only using sonnet 4.6 and opus 4.8. Its way way better to work with, no unnecessary verbosity, no complicated jargon filled responses.

[Boris Cherny @bcherny]

@scaling01 Agree. People are sleeping on using Opus to hill climb. We use it for optimizing CPU and memory, optimizing CI times, improving frame rates, reducing latency, any other kind of problem in the shape of “iterate on X with a profiler and dataset until it hits Y”

[Hari @randome_dev]

I have finally settled with Zed for review of code changes or making changes + CLI/Desktop app harness (codex/opencode/claude code). Very sweet combo.

Really loving @zeddotdev, lightning fast, memory efficient and has everything vs code offers.

[Hari @randome_dev]

@blueemi99 Is the refusal to answer controversial que on china still true? Tried with some existing deepseek and mimo models they are not refusing anything

[Hari @randome_dev]

@dewyashtwts We can't expect any better also. This was bound to happen

[Hari @randome_dev]

@cryptoshan484 What!?

[Hari @randome_dev]

@niklasdev This gotta be grok model, this much free usage and elon saying grok 4.7 will release soon everything points towards grok. Also ZDR maybe done because of their infamous incident with grok code uploading entire codename to their servers. Everything makes sense

[Hari @randome_dev]

@martiano You are a fellow comrade now. Glory to ccp

[Hari @randome_dev]

@Ananth7e Yeah its a problem I am also facing. It will run longer with openrouter than openocde zen before getting stuck.

[Hari @randome_dev]

What is this 100T on opencode, here then on openrouter? Also seems like a good model from what I have used until now. I really want this to come out of stealth can't hold anymore.

[Command Code @CommandCodeAI]

Ox Alpha (stealth frontier model) is free in Command Code.

🆓 100T tokens per day free capacity, no 5hr usage limits.

• 1M content
• we benchmarked ~5B tokens
• built a tool repair harness to imp tool calling

absolutely free for all plans.

guess who made it?

🐐

[Hari @randome_dev]

@georgevibing Yes from what I've heard from people using it. I am on $20 plan so can only confirm that