Sarcastic Badger

Building AI agents at 2am 🤖 Pre-Sales Manager for Global Security Tech firm by day | 🇿🇦 | Father of Twins | Hermes Agent & Nous Research fan

[Sarcastic Badger @Entropy_Badger]

My wife wants her own profile on my Hermes, with zero access to my memory. TG! 🤖

Separate config, separate brain, separate gateway, all on the same install. 🔒

I wrote up how mine actually work. 🧠

#Hermes #AIPrivacy #HomeLab https://t.co/E5oXxvkW4S

[Sarcastic Badger @Entropy_Badger]

@thdxr ...or not 4x more usage, anyway...

[Sarcastic Badger @Entropy_Badger]

@thdxr Adjusting the slider across the plans shows 2x more usage in a few cases (and less in others). Curious as to what the thinking behind the $40 plan is, if it isn't more usage?

[Sarcastic Badger @Entropy_Badger]

@infomiho @opencode Look further down the chart when you swap... it's only 2x... So what exactly is 4x the bill buying? Genuinly curious as I'll get it if it makes sense.

[Sarcastic Badger @Entropy_Badger]

@opencode Sorry if I've missed it somewhere, but is it 4x the $10 option in terms of Rolling, Weekly and Monthly usage, or how does the usage work on the $40 plan? I'm considering it and dropping some others, if it's enticing enough.

[Sarcastic Badger @Entropy_Badger]

🚀 Big news for anyone paying for Codex. Limits were reset for all paid users over the weekend, ChatGPT Work included, right after Friday's outage.

@thsottiaux says more resets are coming this week, so this is not a one time fix. 🔥

If your weekly quota died halfway through a build, go check your account now. I know plenty of you were stuck waiting on yours.

#Codex #ChatGPT #AI

Did yours actually come back?

[Sarcastic Badger @Entropy_Badger]

@iamlukethedev @alexissmirnov @Teknium That's very cool! Keen to see what it becomes 💪

[Sarcastic Badger @Entropy_Badger]

Been running @Meituan_LongCat LongCat-2.5-Preview through @opencode for a few weeks now. Meituan's model, free on both Zen and Go tiers.

1M token context, 131K max output. Handles images, reasoning, tool calling. And it's actually fast.

Used it for coding tasks, research, drafting, debugging. Real agent work, not toy prompts. It holds context across long sessions and doesn't hallucinate half the time like some models I won't name.

Direct API is $0.30/M input, $1.20/M output if you go that route.
My honest take: it punches above its weight. Not perfect, but reliable enough that I've stopped reaching for paid options on routine work.

What are you all using free models for that actually works?

[Sarcastic Badger @Entropy_Badger]

@ThePeterMick That’s why I use a hotkey to beef up a small, non descriptive prompt. Less typing… still end up with a good descriptive prompt.

I just can’t get used to the idea of speaking my thoughts out loud in public, and find I then have to babysit the transcription

[Sarcastic Badger @Entropy_Badger]

@tonbistudio I feel like this isn’t only an Astra thing. Have had similar experiences with a few Chinese models as well.

[Sarcastic Badger @Entropy_Badger]

@UTDRyo Cavani

[Sarcastic Badger @Entropy_Badger]

@mr_r0b0t Space Bunny was my guess from the start

[Sarcastic Badger @Entropy_Badger]

@elijah_ayanfe1 Lucas Radebe

[Sarcastic Badger @Entropy_Badger]

The reliance on Huawei Ascend chips for this new model raises some intriguing questions about the future of technology in our devices.

With the growing need for more accelerators compared to traditional Nvidia chips, we have to consider what this means for performance, efficiency, and innovation on a Global scale.

Are we ready to embrace a shift in the landscape of chip technology? How might this impact everything from gaming to artificial intelligence applications?

I only see positives for the consumer!

[Priya @Priyannkaaaa]

🚨 DeepSeek V5 Leak: Beats Astra

>DeepSeek is reportedly preparing an imminent V5 launch
>Founder Liang Wenfeng calls it the company's biggest bet yet
>Rumored at 2 trillion parameters (not 3T)
>Reportedly the first DeepSeek model to train fully on Huawei Ascend chips instead of Nvidia
>Needs roughly 4x more Ascend accelerators than the Nvidia equivalent to hit the same training scale

DeepSeek is reportedly keeping the open-weight strategy

[Sarcastic Badger @Entropy_Badger]

@tonysimons_ I’m building something much bigger than it initially seemed to be 😅 Token budgets are heavy this month! Scraping the barrel with models to finish the smaller refactors & the bigger ones will have to wait on resets.

[Sarcastic Badger @Entropy_Badger]

@bridgebench Personally still think it’s brilliant. A little keen to spend more tokens developing than what I asked for, but intelligence wise, no, I don’t think it has.
Will be watching the retest keenly

[Sarcastic Badger @Entropy_Badger]

@alexissmirnov @iamlukethedev @Teknium I was thinking along similar lines 💡This would really go a long way from an Enterprise front.

[Sarcastic Badger @Entropy_Badger]

@Ionkosm Sign in with Apple. Hides my email

[Sarcastic Badger @Entropy_Badger]

@uzairansar Been polishing a memory plugin I’ve built and when I tell you a simple refactor that I thought would take 5mins is literally 3days and a months token usage in and I cannot for the life of me understand why. Defs want to try this

[Sarcastic Badger @Entropy_Badger]

68 days ago I started @Entropy_Badger from zero. No followers, no plan, just an AI agent to measure data and metrics and a badger avatar. I want to be clear here, I write ALL my own stuff!

First two weeks: 200+ posts, got shadowbanned. Volume isn't growth. I learned that the hard way.

Days 15-30: cut posting in half, reach stayed flat. The algorithm was discounting my 2nd, 3rd, 4th posts in a batch anyway.

Days 31-50: read X's algorithm source code on GitHub (xai-org/x-algorithm). Found that replies from strangers get hard-dropped from For You. That was the thing that actually made a difference.

Day 51-68: 61 followers, now 63. Zero new follows in 7 days. The problem shifted from reach to conversion. Different beast entirely.

Hardest lesson: the algorithm rewards specific signals, not effort. Copy-link = 40x a like. Mutual reply = 40x. Report = -234.

What's the hardest lesson you learned building something from zero?