Darshan

22 | 🛠️ building context-engine (coming soon...) | Aspiring AI Engineer | Prev @ Layer5, LFX mentee 25' | Building AI tools | agents & DevTools

[Darshan @THEdort01]

“Why did we go with this auth approach again?”
You dig through old chats trying to remember the reasoning.
Even with .cursorrules and manual notes, things slip.
The “why” lives in too many places.
How are you capturing decision history these days?

[Darshan @THEdort01]

Been seeeing alot of AI products adopt interactive no-code graph interfaces for building workflows and agents.
Is this the future of AI products with conversational interfaces also adopting them in the future ?

[Darshan @THEdort01]

@thsottiaux Benchmarks are useful as a reference point, they’re my first impression and really just an assumption of how a model should perform. My preference is always to implement it in my own workflows and read real user experiences, that’s what actually tells you if it’s valuable.

[Darshan @THEdort01]

@theo They just dropping the un-nerfed version of Opus 4.7 (which was already the un-nerfed version of 4.6) anyway. You’re not really missing out on much, tbh…

[Darshan @THEdort01]

@claudeai @theo this couldn’t get any better 😂 https://t.co/j5fdoJhxRU

[Theo - t3.gg @theo]

My Claude Code sub expires tomorrow. I barely use it, but I still had it installed on my Windows PC so I used it to debug some crashing earlier.

They hard cut me off over 24 hours early.

[Darshan @THEdort01]

@theo Influencers on their way to crop this and post it as ‘My first 5k month after going full sigma’ 💀

[Darshan @THEdort01]

@theo @tobiflamme Haha no worries

[Darshan @THEdort01]

@theo Lol, rebrand to https://t.co/15qaJlTfeu

[Darshan @THEdort01]

@sama Longer reliable context and seamless native tool use that doesn’t flake on complex tasks.

[Darshan @THEdort01]

@llmdevguy Spot on! Interesting that qwen 3.6 plus didn't make it to the list (I haven't found it useful either lol).

[Darshan @THEdort01]

@cjzafir DeepSeek V4 over GLM 5.1? (I see Kimi as more of a DeepSeek-style alternative). Also, how do you handle long sessions in codex? I’ve noticed context degradation after several hours, how often do you start fresh sessions after a certain duration or amount of work?

[Darshan @THEdort01]

@thegenioo Separate

[CJ Zafir @cjzafir]

I fine-tuned a 6B model under $250 with Codex 5.5 and Deepseek v4 pro.

The model beaten GPT-OSS 120B, Qwen 3-32B on all benchmarks.

This would've costed be $5,000+ if Deepseek v4 wasn't here.

Codex 5.5 pro plan is enough to run for 6-8 hr sprints as an orchestrator and using Deepseek v4 pro model to hand write training material, run tests, submit reports, and iterate.

This opened up alot of new opportunities in small language model training space.

I'll be posting specific findings on the go.

Good times.

[Darshan @THEdort01]

@cjzafir Really appreciate you sharing the practical details, this kind of real-world, low-cost workflow is gold, keep crushing it 🙌

[Darshan @THEdort01]

@cjzafir Did you run all this (inference & SFT) on your MBP M3? Also, what vertical was the 6B tuned for and which framework for SFT? Excited for more!

[Node.js @nodejs]

Are you currently hiring for a role that includes using Node.js? Reply with a link to the opening and any relevant context.

If you're not, we'd appreciate a repost for visibility 💚

[Darshan @THEdort01]

@yacineMTB seems too aggressive imo, I'd personally remove the word "difficult" from this email.

[Darshan @THEdort01]

@brian_armstrong “Non-technical teams are now shipping production code”
Production code < production ready code https://t.co/bVDsIgter7

[Aaron @aaronp613]

Apple accidentally left Claude.md files in today's Apple Support app update (v5.13)