Ryan

[Ryan @HumanHitsSend]

Although this is not a problem for me any more, I’m glad someone plugged this hole.

[Ramit @RamitWasTaken]

I keep two Claude Code accounts, personal and work. Switching meant signing out, signing back in, and retyping the whole login.

Built Janus. One click from the menu bar, plus how much of each account's 5-hour and weekly limit is left.

Free, MIT, macOS. https://t.co/v7roRzAioF

[Ryan @HumanHitsSend]

I love that we keep pushing the envelope with these open models. Exciting to see what gets cooked up in such a short time! I have no doubts that trillion parameter models will one day fit in your Apple Watch 😅😅😅.

The @NVIDIAAI DGX Spark box is such a blessing. Prices are going up. Retail is hitting $6500 as of this AM. This could still feel cheap in the future.

[mr_r0b0t @mr_r0b0t]

r0b0tlab/GLM-5.3-Flash-EXL3-2.25bpw-sm121 + 3.00bpw DFlash2 drafter + native ExLlamaV3 runtime!

Tested on a single @NVIDIAAI
GB10 ♥️

At 8K C1, code: 22.61→49.96 tok/s with K=5
Exact single-key retrieval at 259,993 prompt tokens
Extended eval suite in progress 🤓

[Mia @MiaAI_lab]

TensorFold is going to change everything 🚀

Exciting times ahead! ⏳

[Ryan @HumanHitsSend]

@CodexResets1 These guys are trying so hard to keep up. Nice watching Altman’s “non profit” power play burning to the ground.

OpenAI is such a follower - they literally had this opportunity before anyone

[Jun Song @jun_song]

Honestly, you don't need a college degree anymore.

AI is moving faster than any education.

[Ryan @HumanHitsSend]

@callebtc I found Codex to be very inconsistent in every way you wouldn’t want a paid model to be inconsistent.

[Ryan @HumanHitsSend]

I cancelled OpenAI 2 months ago. I found Sol to be a waste of time, the model was very inconsistent and let’s face it, Opus 5.5 is the most balanced and best model out there now.

Now with 3 DGX sparks, I’m using only 1 Claude Max + Grok … and Grok is on the chopping block with my locals getting better and better.

[calle @callebtc]

heavy codex user here.

$200 plan used to be incredible value. then, openai got so greedy that it now barely lasts 2 days anymore.

got a $200 claude plan out of frustration.

turns out opus 5.5 is amazing and feels like 5x more usage than what i get with astra.

[Ryan @HumanHitsSend]

If you don’t like open models, you’re doing it wrong

[Mia @MiaAI_lab]

Useless in what?

For being completely private?
For having no limits?
For not being nerfed after a while?
For having no downtime?

For being able to own intelligence instead of renting one?

Yes, without all of these local models are completely useless I agree 💯

[rari @0xwhrrari]

Jev Founder, Diogo Almeida, just released a 12-page PDF on how to use Jev with LLMs

It is more useful than most paid AI courses:

this is a 10-step blueprint on how to build a faster, cheaper and more controllable AI system around Claude, Codex, Grok or any other LLM:

step 1 → split the responsibilities: the LLM generates, Jev makes bounded semantic decisions and deterministic code keeps authority

step 2 → build the state: give Jev the current request, relevant evidence, policy and proposed action instead of sending the entire conversation

step 3 → choose the right primitive: Choice selects a route, Score evaluates an ordered rubric and Noul returns the probability that a statement is true

step 4 → replace giant evaluation prompts with atomic questions: intent, urgency, evidence, risk and scope become separate typed decisions

step 5 → put Jev before the LLM: select the context, tools, provider and workflow before paying for an expensive generative call

step 6 → give the LLM a bounded job: once Jev selects the route, the model receives only the instructions, files and tools required for that branch

step 7 → put Jev after the LLM: check whether the result answers the request, uses sufficient evidence and stays inside the permitted scope

step 8 → route by confidence: high-confidence low-risk cases proceed automatically, uncertain cases request more context and consequential actions go to review

step 9 → batch independent decisions: ask multiple Choice, Score and Noul questions over one shared state instead of creating another LLM call for every judgment

step 10 → record the complete decision receipt: state version, question, probabilities, selected route, model, latency, outcome and human override

most AI courses teach you how to write a bigger prompt

this 12-page guide teaches you how to build the control system around every prompt

the result: smaller contexts, fewer unnecessary LLM calls, safer tool execution and decisions you can actually inspect, test and improve

Send this PDF and the original Jev article to Claude Code or Codex and start rebuilding one expensive LLM decision at a time ↓

[Mia @MiaAI_lab]

Many asked for this. You can now browse all 100 of my HTML experiments in one place.

This page will continue to be updated as I add more experiments in the future.

https://t.co/mJPbuV0SUa

[Ryan @HumanHitsSend]

Mia always on top with the fire 🔥 recipes.

[Mia @MiaAI_lab]

A good day for updates ⚡️

Update for Qwen3.8 Flash for a solo DGX Spark

- Decode on prose is now 49 tok/s single stream.
- Decode on code is now 62 tok/s single stream.
- Faster follow-up replies.
- Optional vLLM 0.30 path with official Nvidia NVFP4.

This is still the BEST model to run a single DGX Spark.

[Martti Malmi @marttimalmi]

Nostr VPN is a decentralized Tailscale alternative that needs no email addresses, user accounts or control servers. Now it also lets you buy and sell VPN bandwidth for bitcoin. https://t.co/QUBFVZ2Bag

[Ryan @HumanHitsSend]

@mtsmachado8 @BuildwithOmkarr I have mine fully self hosted on my server. It can run on my iOS app or the desktop app from OMB that I’ve customized to my liking. It’s a great OS repo

[Ryan @HumanHitsSend]

@mtsmachado8 @BuildwithOmkarr Just take the stuff that works and make it your own. I’ve got a fully custom build with locals, self hosted, my own iOS app. I worked around a lot of the bugs and goofy stuff with permissions and shell asking a thousand times even when you click “allow for session”

[Ryan @HumanHitsSend]

@BuildwithOmkarr I’ve been building with OMB and have a really great set of custom teams. I’ve tweaked it to run only my own local models and also my own agent computer (not the cloud sub 😏).

Thanks for building this. It is a great repo with lots of exciting opportunity

[Omkar Satpute @BuildwithOmkarr]

🎉 Introducing OpenMausBot

An open-source, self-hostable team of AI agents, not another boxed-in assistant.

• Computer use: browser, terminal, files & a real desktop
• Connectors for the apps you already live in
• Goals, routines & progress that keep moving
• Built for mobile and web

Works with @claudeai , @ChatGPT , @grok , @cursor_ai , @NousResearch , @Alibaba_Qwen , @Kimi_Moonshot , and any harness or model you have.

Your bots. Your machine. Your keys.

Repo → https://t.co/fqLTS7CFhs

[Ryan @HumanHitsSend]

@TheGeneral_0 Jelly roll calling anyone fat is the richest stuff online

[Ryan @HumanHitsSend]

@cmitsakis @jun_song the degree is already obsolete, at least you get 2 years here

[huihui.ai @support_huihui]

Found this project through a search; we haven't tried it.

https://t.co/vInTllq5yc

[HelenaChase @HelenaChas61775]

So... the government and organizations they want to have access, do have it. But your subscribers, developers and businesses still do not?

Looks like you made a very powerful AI system for the government to use however they see fit whether you like it or not.

Your public refusal to have your systems used as weapons, and then your public circus about how dangerous the weapon you made is, clearly got some attention.

Was that the plan all along? Or did you really screw up this badly?