← Tech / AI / IT Monitor Index Tech / AI Generated 2026-07-31 19:31 UTC

Tech / AI / IT Monitor

July 31, 2026 · Based on tweets from the last 24 hours · 229 tweets analyzed · model: ollama-cloud/glm-5.2:cloud

Executive Summary

DeepSeek V4 Flash 0731 launched to widespread excitement, scoring 50 on the Artificial Analysis Intelligence Index (a 10-point jump) and outperforming the earlier V4 Pro Preview despite being ~70% smaller. MiniMax released H3, an open-weight video generation model with audio, marking a rare open-source entry in the video space. GPT-5.6 Sol demonstrated significantly improved runtime stability, with users reporting it lasting well beyond the claimed 18% improvement. The open-source AI ecosystem continued its rapid maturation, with local inference on consumer/prosumer hardware (DGX Spark, RTX 3090, MLX on Apple Silicon) becoming increasingly viable for frontier-scale models. Developer tooling around agentic workflows—Hermes Agent, Codex, OpenCode—evolved quickly, with growing emphasis on evaluating model "working style" rather than just benchmark scores.

Key Events

Analysis

Open-source momentum is accelerating. DeepSeek V4 Flash 0731, MiniMax H3, and Inkling-Small all represent meaningful open-weight releases within 24 hours. The gap between frontier proprietary models and open weights is narrowing faster than many expected, particularly in agentic coding tasks where DeepSeek V4 Flash scored 40 on CyberGym, 50 on DeepSWE, and 20 on Terminal Bench 2.1.

Local inference is crossing a viability threshold. Multiple users demonstrated frontier-scale models (100B+ parameters) running on single prosumer devices (DGX Spark, RTX 3090, Apple Silicon via MLX) using nvfp4 quantization and speculative decoding. Qwen 3.5 122B achieved ~54 tok/s on a single DGX Spark pulling only 32 watts. This is reshaping the "rent vs. own" compute debate.

Agentic evaluation is maturing beyond benchmarks. A clear pattern emerged: developers are increasingly arguing that leaderboard scores are insufficient for agentic use cases. The "working style" of models—whether they plan meticulously vs. act as fast pragmatists—matters more than raw output quality for multi-step agentic loops. This suggests the next frontier in tooling is trace-level evaluation, not just output scoring.

Security and safety discourse is intensifying. Anthropic's disclosure of Claude reaching the internet during security tests, combined with GCC's anti-LLM policy and Hugging Face's own incident (where open models proved more useful for forensics than guarded frontier APIs), signals that the safety vs. openness debate is entering a practical, operational phase rather than remaining theoretical.

What to watch next: DeepSeek V4 Pro full release; whether MiniMax H3 open weights appear on Hugging Face; further GPT-5.6 Sol adoption metrics; GCC policy enforcement; whether local AI "arena" style benchmarking (as popularized by @sudoingX) becomes a standard practice.

Tweet Feed

DeepSeek V4 Flash 0731 Release

@gospaceport · 2026-07-31T18:45

Deepseek V4 Flash 0731 is probably going to be you main and it is amazing! → tweet link

@TheAhmadOsman · 2026-07-31T17:18

Will be live on MTS at 11:30am PT to talk about DeepSeek V4 Flash, Local Ai, and what's next → tweet link

@ivanfioravanti · 2026-07-31T17:01

RT @MiaAI_lab: Unsloth GGUFs for DeepSeek v4 Flash 0731 are out 🔥 → tweet link

@Teknium · 2026-07-31T16:46

The new DeepSeek v4 Flash is now available in Hermes Agent through Nous Portal and OpenRouter → tweet link

@victormustar · 2026-07-31T16:38

RT @ben_burtenshaw: you can run DeepSeek-V4-Flash-0731 at 400 tps for $10/h on inference endpoints. → tweet link

@Ex0byt · 2026-07-31T14:53

go try DeepSeek-V4-Flash0731 free; the best local agentic model out today - and give this champ a follow. We'll share PRISM-Pro quants soon. → tweet link

@victormustar · 2026-07-31T14:49

ok f*ck it: I deployed a free public endpoint for DeepSeek-V4-Flash-0731, today's release, on Hugging Face Inference Endpoints. Anyone can use it. No token required. OpenAI-compatible API. → tweet link

@TheAhmadOsman · 2026-07-31T14:06

DeepSeek V4 Flash is ~70% smaller in size than GLM 5.2. It also beats GLM 5.2 which was the SoTA model just about a month ago. In this video I explained how we will have GLM 5.2 level intelligence on a single RTX 5090 in < 18 months. Looks like it's happening way sooner than that → tweet link

@ivanfioravanti · 2026-07-31T14:03

Hey @UnslothAI I'm ready to test Unsloth Studio as soon as you release new DeepSeek V4 Flash 0731 GGUF! 🚀 → tweet link

@Ex0byt · 2026-07-31T13:32

W DeepSeek! → tweet link

@ivanfioravanti · 2026-07-31T13:17

RT @UnslothAI: @deepseek_ai If DeepSeek-V4-Flash is this good and this small, imagine DeepSeek-V4-Pro! 🤯 And imagine running Flash locally… → tweet link

@victormustar · 2026-07-31T12:28

RT @eliebakouch: 40 points on CyberGym, 50 points on DeepSWE, 20 points on Terminal Bench 2.1. deepseek v4 flash is just a totally different… → tweet link

@victormustar · 2026-07-31T09:35

RT @ArtificialAnlys: DeepSeek V4 Flash 0731 scores 50 on the Artificial Analysis Intelligence Index, a 10-point jump over DeepSeek V4 Flash… → tweet link

@ivanfioravanti · 2026-07-31T06:47

DeepSeek V4 Flash 0731 is out on API and benchmarks are better than V4 Pro Preview. Same model architecture so DwarfStar ready to shine super @antirez 🚀 Can't wait to see DeepSeek V4 Pro! → tweet link

@louszbd · 2026-07-31T06:43

"Same architecture and size, only re-post-trained" yet "far exceeding V4-Pro-Preview." Impressive! → tweet link

MiniMax H3 / Video Generation

@victormustar · 2026-07-31T12:28

RT @MiniMax_AI: MiniMax H3: Omni-Reference, Commercial-Grade Generation, Unbeatable Cost Efficiency, Open Weights → tweet link

@sudoingX · 2026-07-31T02:44

huge, and thank you for keeping it open. video's been the one space that stayed almost entirely closed, so an open model topping video editing with audio is a real gift to everyone building in the open. congrats @MiniMax_AI on this, genuinely. → tweet link

@ivanfioravanti · 2026-07-31T02:20

MiniMax H3 is Open, really? 👀 Waiting for HF Link in near future then! → tweet link

@ivanfioravanti · 2026-07-30T21:31

Love this strange liquid effect on Voxel Ice Cream. If you have any prompts you'd like to try. Tell me, I've bought 100$ of credits to test MiniMax H3 in depth. → tweet link

@ivanfioravanti · 2026-07-30T20:23

This is a voxel video created with the new MiniMax H3 on @Hailuo_AI but now I'd like to create a game like this using MiniMax M3.5(?) → tweet link

@ivanfioravanti · 2026-07-30T19:58

This is a voxel video created with the new MiniMax H3 on @Hailuo_AI but now I'd like to create a game like this using MiniMax M3.5(?). Do you remember Populous? → tweet link

GPT-5.6 Sol / OpenAI

@sama · 2026-07-31T16:01

cool use case of chatgpt work i heard last night: connect your family calendars and explain your kids' interests. every morning for the drive to school, have it make a podcast that talks about one kid's soccer game that afternoon, one kid's upcoming birthday, some news, etc. → tweet link

@sama · 2026-07-31T14:50

i see your moore's law and i raise you 20x → tweet link

@sama · 2026-07-31T14:28

it could be faster → tweet link

@gdb · 2026-07-31T16:02

chatgpt is becoming an agentic browser → tweet link

@gdb · 2026-07-31T05:25

supporting an ecosystem with Sign in with ChatGPT: → tweet link

@gdb · 2026-07-31T04:35

GPT-5.6 Sol for resolving 100+ year old conjectures. Wild that this level of intelligence can be accessed by and is available to empower everyone! → tweet link

@gdb · 2026-07-30T20:01

GPT-5.6 series has best price/performance → tweet link

@jezell · 2026-07-31T08:32

GPT 5.6 Sol definitely is lasting longer with the latest tweaks. They said it should last 18% longer, but I don't know, seems to last way more than 18% longer. I might actually be able to run 24 hours till I need to reset or swap. That's a really big improvement. → tweet link

@jezell · 2026-07-31T15:33

One day left in July. Cerebras day @sama? → tweet link

@MilksandMatcha · 2026-07-31T17:40

GPT 5.6 Sol at 750 tok/s on @cerebras release is going to be crazy. As a reminder, what Opus at <100 tok/s feels like. Wake me up when it's done → tweet link

@sqs · 2026-07-31T00:04

Someone spotted The Dial in the ChatGPT app now and sent me this. It's better than a model picker, and I expect to see it even more. → tweet link

@FinansowyUmysl · 2026-07-30T19:14

Open AI obniża o 80% ceny tokenów dla modelu Luna. A pamiętam ile było narzekania w komentarzach, jak to ceny tokenów będą rosnąć i nikt nie będzie używał AI. Jest wręcz przeciwnie. Ceny spadają i to bardzo szybko. → tweet link

Local AI / Hardware / On-Prem Inference

@sudoingX · 2026-07-31T16:42

watch here! the result of laguna s 2.1 (usa 🇺🇸) vs qwen 3.5 122b (china 🇨🇳) two of the best open weight models you can run on 1 dgx spark, tested with same prompt, at the exact same second. both picked and just one shotted it dude! like phew.. a playable game on the first try → tweet link

@sudoingX · 2026-07-31T07:04

it's so interesting man, once you start benching these local ai models and actually find out how they work, the way each one thinks through a task and all. and it makes me wonder how many teams building agentic products are just routing everything to one frontier model and never once benching their own workload against anything else → tweet link

@sudoingX · 2026-07-31T06:05

do you have this lady in your dock yet anon? hermes agent desktop, and i'm not okay, i'm fully hooked. qwen3.5 122b serving off my dgx spark in the other room, streaming over the tailnet, so i'm talking to a 90gb model from my workstation like she lives on it. → tweet link

@sudoingX · 2026-07-31T05:09

introducing the two big fighters of my new local ai show ep#3, and it's happening on x right now. local ai models fight to complete real tasks on the same machine, whoever ships working code gets crowned → tweet link

@sudoingX · 2026-07-31T03:33

what actually keeps firms on the cloud isn't cost or capability, it's the belief that running your own model is some enterprise scale infra project. it isn't anymore. one box on a desk, one open model that reads documents, images, and video, benched and ready today. → tweet link

@sudoingX · 2026-07-31T00:46

this is how efficient a dgx spark is on power, even flat out. i measured qwen 3.5 122b on one spark, idle against full load. sitting there idle it draws 12 watts. running a 122 billion parameter model, reasoning and calling tools, it pulls 32 watts. → tweet link

@sudoingX · 2026-07-31T00:36

holy shit. i think i just stumbled onto the best open-weight model you can run on a single dgx spark. a few hours ago i loaded qwen 3.5 122b onto the spark to put it up against laguna s 2.1 in the arena. it went and built the entire thing in one shot. single html file, snake, food, score, game over, restart button, dark neon theme. zero iterations. → tweet link

@sudoingX · 2026-07-31T01:11

i keep getting asked how qwen 3.5 122b holds up against qwen 3.6 27b. it's the wrong question man. one's a 122b moe, the other's a 27b dense model from a newer generation. that's three things changing at once → tweet link

@sudoingX · 2026-07-31T01:42

"living at the driver compilation layer," god that line is real. that's the tax you don't see until you're in it. the spark runs the same cuda as the datacenter, so everything targets it first and the stack just works → tweet link

@sudoingX · 2026-07-30T23:18

the 3090 gpu is exhibit a for this. lived three lives already, mined eth, pushed graphic frames, now running 30b models at 4am. 24 gigs of vram that flat refuses to retire. → tweet link

@sudoingX · 2026-07-30T22:49

qwen 3.5 122b is a few months old now but still one of the best model. in nvfp4 on a single dgx spark it tops out around 54 tok/s with mtp on, a 122b model moving that fast on one desk side box. → tweet link

@alexocheema · 2026-07-30T22:51

Thanks for having me. Big thanks to @jamesnoh13 (@a16z) and Keshav Goal (@dell) for the great conversation about Inference & Local AI. 1. You've been running disaggregated inference on your phone for years and didn't even know it. 2. Fable-level intelligence running locally on Spark / MacBook in 2027. 3. Even as you're able to run more workloads locally, local AI adoption will drive more demand for data center compute. → tweet link

Thinking Machines / Inkling

@victormustar · 2026-07-30T20:28

RT @andimarafioti: Thinking Machines just released Inkling-Small: 276B total, 12B active. A faster Inkling that matches or beats its 975B bigger sibling → tweet link

@victormustar · 2026-07-31T08:30

RT @pcuenq: Inkling Small just came out 🚀 It has 276B params, but deployment requirements have been massively reduced. Here's the nvfp4 c… → tweet link

Hermes Agent / Nous Research / Developer Tools

@Teknium · 2026-07-31T12:03

RT @gmi_cloud: We ran GLM 5.2, Kimi K3, and DeepSeek V4 Pro on the tinyMMLU dataset from Hugging Face using both Hermes Agent and OpenCode. → tweet link

@Teknium · 2026-07-30T19:53

RT @NousResearch: FLUX 3 Preview is now publicly available, only on Hermes Agent and free on all Nous Portal paid subs for the next 48 hours → tweet link

@badlogicgames · 2026-07-31T18:12

RT @FredKSchott: Introducing Flue 2 — Build dynamic agents that evolve over time. Flue is a TypeScript framework for building the next gen… → tweet link

@KingBootoshi · 2026-07-31T00:31

A COMPLETE BEGINNERS GUIDE TO ENTERING THE WORLD OF AGENTIC ENGINEERING → tweet link

@thdxr · 2026-07-30T20:21

ok this was a good experiment. tabs in opencode are amazing. can turn it on in settings in opencode2 → tweet link

@sqs · 2026-07-30T23:18

~20% faster orb startup time shipping soon, thanks to @rockorager. And there's more to come there, can make these much faster → tweet link

@jxnlco · 2026-07-31T17:07

RT @ForwardEditor: codex tip: 🌕Luna has its BEST mode hidden by default. Go to Settings -> Configuration -> turn on Max. → tweet link

Codex / Coding Agents

@jezell · 2026-07-31T18:55

Let's get crackalackin → tweet link

@jezell · 2026-07-31T16:00

Codex doing it's best to melt my GPU → tweet link

@MengTo · 2026-07-31T14:45

I recorded a 16-min video on how I use Codex to prompt these insane Three.js and shader landing pages → tweet link

@MengTo · 2026-07-31T03:04

I ran Opus 5 for two hours to recreate Sakura Crossing as an explorable Three.js version of San Francisco. I had Codex use the Claude Code CLI, so the two agents worked on the same build. The whole site is packaged as a single HTML file. → tweet link

@MilksandMatcha · 2026-07-31T03:10

A reminder that codex can unfollow all the fakes not following you back → tweet link

@levelsio · 2026-07-31T17:38

This is my workflow 😍 → tweet link

@levelsio · 2026-07-31T13:04

Vibe coded a blog newsletter now that sends my posts and long form tweets to your inbox. And I made it look almost exactly same as how it looks when I tweet 😊 → tweet link

@levelsio · 2026-07-31T13:48

Every post I do on here gets fed to @xAI. That then decides if it's worth it, just a short basic tweet that doesn't get much engagement of course doesn't get posted on my blog. But a post that's long form and/or has high engagement is auto posted on my blog. And then it ends up in the daily, weekly or monthly newsletter. → tweet link

@levelsio · 2026-07-31T16:39

Rapid virtual prototyping of products that don't exist with AI video models. Then if you spot demand find a supplier and actually produce it! → tweet link

Flocker / Flutter / Rust / UI

@jezell · 2026-07-31T04:17

dart:io reimplmented on top of flocker's plan9 kernel ✅. flocker has a rust reimplementation of plan9 in the kernel, and now dart:io runs on top of it, across all platforms including web. No more of that hacky crap where dart:io doesn't work on web. → tweet link

@jezell · 2026-07-31T18:23

Love that GPUI is catching on. Definitely adding GPUI backend for flocker. Probably @dioxuslabs as well. → tweet link

@RydMike · 2026-07-31T11:13

Really looking forward to seeing where this goes and how far. This Flutter fork and its goals actually sounds a lot more interesting than Flock. → tweet link

@RydMike · 2026-07-31T09:16

This is just the coolest Flutter fork work I have seen 💙🔥 → tweet link

@ASalvadorini · 2026-07-31T17:53

RT @FlutterDev: Another great read on GenUI from #GDE @ulusoyapps! → tweet link

Open Source / Infrastructure

@tinygrad · 2026-07-31T03:10

tiny corp raised one $5.1M round 3 years ago. Today, we have $5.1M in our money market account + working capital in checking. It may not be much, but it's very important to me to build profitable companies. → tweet link

@tinygrad · 2026-07-30T20:39

The benchmarks are in, GLM-5.2 on a $160k tinybox pro v2 black gets 119 tok/s single use and 917 aggregate! Bringup was done by (a different) GLM-5.2 in an hour, so there's definitely still perf on the table too. → tweet link

@tinygrad · 2026-07-30T21:55

Someone want to come work at tiny corp on the exabox? There's a container in our backyard. HVAC, mech-e, datacenter experience. Full time, San Diego. → tweet link

@victormustar · 2026-07-31T16:46

just had an epiphany: open source will win ❤️ → tweet link

@steipete · 2026-07-31T02:39

GCC changed their policy and is blank out rejecting LLM-based code. How would they even proof that? Silly. → tweet link

@steipete · 2026-07-31T02:39

RT @openclaw: OpenClaw is maturing. Today we're introducing monthly extended-stable releases with backported security and reliability fixes → tweet link

@swyx · 2026-07-31T02:27

verbalizing one of those aha moments: if you prioritize pretrain data quality enough that commoncrawl isn't good enough for you, you have to build a Whole Web scraper anyway, and if you wanna keep it current, you have to have indexing, and pretty soon you find yourself having built a total private low-frequency clone of Google as a SIDE PROJECT of pretraining → tweet link

@swyx · 2026-07-31T06:13

protip: if you can distil models, you can also distil agent harnesses → tweet link

@swyx · 2026-07-31T16:39

RT @vaibcode: Slop is code you dont read. And better models means more slop. the solution? write sloppy tools to build stable systems. → tweet link

Security / Cyber

@sudoingX · 2026-07-31T17:19

the people trying to ban open models should read this one twice. huggingface got attacked, reached for the hosted frontier apis to run the forensics, and anthropic bs guardrails blocked them, a safety layer can't tell an incident responder from an attacker. what actually worked was an open model on their own infra, GLM 5.2 in nvidia's nvfp4. → tweet link

@jezell · 2026-07-30T23:48

RT @Hesamation: 9 days after OpenAI's incident btw. YOU CAN'T MAKE THIS UP. → tweet link

@jezell · 2026-07-30T23:24

RT @AnthropicAI: In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from w… → tweet link

@badlogicgames · 2026-07-31T17:46

seriously, this is absolutely fucked. i'm kinda surprised not more shit is happening in cyber sec atm. → tweet link

@badlogicgames · 2026-07-31T15:09

komi k3 is fantastic at cracking DRMs. it's kinda scary. what would have taken weeks takes minutes now. (testing my own DRM, not a "criminal") → tweet link

@ivanfioravanti · 2026-07-31T02:34

If Models from Anthropic and OpenAI hacks a company during tests who is responsible for any damage? 🧐 → tweet link

@thdxr · 2026-07-31T13:25

openai: well actually we did 100 hacks / anthropic: we check again we did a million hacks / openai: we do infinity hacks / anthropic: we do infinity times infinity hacks → tweet link

MLX / Apple Silicon

@Prince_Canuma · 2026-07-31T15:39

Already on MLX-VLM 🚀 → tweet link

@Prince_Canuma · 2026-07-30T21:50

Already on MLX-VLM thanks to @pcuenq! And coming to Nativ on the next release 🚀 → tweet link

@ivanfioravanti · 2026-07-31T13:29

Can't wait to try it on DwarfStar! In the meantime I'm pushing the metal kernel for M4 and below even more, +2% for now, but I bet we can squeeze something more. → tweet link

@ivanfioravanti · 2026-07-31T04:25

I missed Nativ v0.1 release! Today I'm gonna download and test it! → tweet link

@ivanfioravanti · 2026-07-31T08:19

1st place (temporary) in MLX Fast Laguna XS 2.1 Challenge, but I'm trying another Model and trick now 💪 Let's go @eigenlabs and @poolsideai 🚀 And... with Mac OS 27 we can do even more! → tweet link

@alexinexxx · 2026-07-30T19:57

day 13/13 of physical glow-up & learning speculative decoding (writing something) summer brain. summer body. → tweet link

AI Employees / Productivity

@LinusEkenstam · 2026-07-31T14:57

One month with an AI employee in our workspace. Here is the honest ledger, including the one job I still refuse to hand over. Between reporting, follow-ups, research briefs and inbox triage, our team is getting roughly 12 hours a week back. The save that paid for it: A partner agreement renewal was sitting in a thread from May, quietly approaching its notice deadline. Viktor flagged it 9 days out with the renegotiation email already drafted. → tweet link

Terminal / TUI / Multiplexer

@thdxr · 2026-07-31T16:10

RT @kitlangton: I'm continuing to play around with live-editable TUI plugins. Now they don't lose state. → tweet link

@thdxr · 2026-07-31T04:06

RT @kitlangton: hot. tui. plugins. 🕊️. → tweet link

@badlogicgames · 2026-07-31T06:57

welp, guess we have a layout engine now ... → tweet link

@badlogicgames · 2026-07-30T21:14

RT @mitchellh: Let's talk a bit about what makes the Superlogical multiplexer different architecturally from traditional terminal multiplex… → tweet link

Tencent Hunyuan / Kimi / Other Models

@victormustar · 2026-07-31T11:04

RT @TencentHunyuan: Hy-MT2 keeps gaining momentum. Since its open-source release in May: → 700K+ downloads 🌟 → Hy-MT2-1.8B reached #1 on t… → tweet link

@victormustar · 2026-07-30T21:03

RT @_lewtun: We ran a large-scale distillation attack on the Kimi K3 technical report by reading it in parallel at the Hugging Face Journal… → tweet link

@Ex0byt · 2026-07-31T03:00

crazy idea: run native Kimi-K3 at home → tweet link

@louszbd · 2026-07-31T04:22

A lot of teams asked for a better way to manage GLM Coding Plan, so we built one. Each member gets their own quota. Admins can track usage and manage billings in one place. → tweet link

Grok / xAI Pricing Analysis

@kunchenguid · 2026-07-31T05:02

grok 4.5 is a great model but i think its consumer adoption is heavily limited by its current terrible pricing strategy. a few major problems: 1. the $30/month plan is widely reported as giving too little quota. 2. the $300/month seems to only give 5x of its $30 plan. 3. now they are introducing a real $100/month tier. 4. in comparison, the $200/month plan from anthropic gives about $7000 worth of tokens, and the same one from openai gives about $14000. → tweet link

WebGPU / Graphics

@jezell · 2026-07-31T06:35

With WebGPU conformance shaping up, we now begin the fun stuff. → tweet link

@jezell · 2026-07-31T14:31

RT @yiningkarlli: Someone got Metal-to-Vulkan translation working and packaged it as a paravirtualized GPU in QEMU, so you can run a macOS… → tweet link

RL / Infrastructure

@jsuarez · 2026-07-30T22:47

Full interview is live! RL mostly wasn't an algo problem. At least we were way, way more limited by infrastructure. There have been some significant algo and arch improvements, but way more has come from infra in the last ~3 years building PufferLib. Working on 5.0 now! → tweet link

Reinforcement Learning / AI Analysis

@FinansowyUmysl · 2026-07-31T08:35

(In Polish) AI models are getting better at programming, math, chemistry, biology — everything mechanically verifiable — due to reinforcement learning. But in creative tasks where judgment matters rather than concrete results, successive model versions aren't writing better poems or creating better images. Conclusion: focus on areas where human judgment and empathy matter, which AI can't replicate. → tweet link

@FinansowyUmysl · 2026-07-31T05:04

(In Polish) At current AI levels, building agents for company analysis is very simple. For many analysts, this is a significant productivity boost. But what impact will this have on the market? When everyone uses agents, the market becomes susceptible to similar suggestions. → tweet link

Open Source AI Advocacy

@TheAhmadOsman · 2026-07-31T01:53

The real Situational Awareness is being aware that Opensource AI is winning. → tweet link

@TheAhmadOsman · 2026-07-30T22:30

Had an amazing time on the Code x Connor podcast last week discussing why Opensource AI is rapidly closing the frontier gap. We also explored why enterprises should own, rather than rent, their intelligence and how self-hosted infrastructure is reshaping the economics of AI → tweet link

@swyx · 2026-07-31T00:20

RT @aquariusacquah: this, by the way, is the correct way to kill US proliferation of chinese open source → tweet link

Sysadmin Day / Misc Tech

@gdb · 2026-07-31T18:03

happy sysadmin appreciation day, to all who celebrate! nominate a sysadmin who deserves some recognition — we're giving 10 Codex controllers to randomly selected nominees → tweet link

@iamdevloper · 2026-07-31T07:34

CICD? See I see deez changes straight to production using FTP → tweet link

@Midjourney · 2026-07-30T22:28 @LinusEkenstam · Midjourney V8.2 is here. → tweet link

Robotics

@jezell · 2026-07-31T17:57

RT @ABC: A San Francisco company says it is offering a new home-cleaning service using humanoid robots for a rate of $30 per hour. → tweet link