← Tech / AI / IT Monitor Index Tech / AI Generated 2026-09-12 19:12 UTC

Tech / AI / IT Monitor

September 12, 2026 · Based on tweets from the last 24 hours · 154 tweets analyzed · model: ollama-cloud/glm-5.2:cloud

Executive Summary

The AI industry is currently dominated by a major debate over AI safety and regulatory pacing, sparked by Dario Amodei's proposal to "pace the frontier," which has received public agreement from OpenAI's Sam Altman. Meanwhile, developers are increasingly pushing back against the high costs and "spiky" performance of new closed-source frontier models like GPT-6 Astra and Claude Opus 5, leading to a surge in local model usage on Apple Silicon and DGX hardware. Open-source agent frameworks like Hermes Agent and new dev tools like Cursor Projects are rapidly maturing, shifting the focus from raw model capability to orchestration and autonomous software lifecycle management.

Key Events

Analysis

There is a growing bifurcation in the AI market: frontier labs are urging regulatory capture and safety pauses, while the developer community is heavily resisting, citing risks of "neo-feudalism" where only top companies can compete. Meanwhile, practical developer experience is hitting a wall with closed models—cost-to-quality ratios are declining, and benchmarks are increasingly disconnected from real-world utility. This is driving a massive renaissance in local hardware (DGX Spark, Apple Silicon M5/M6) and open-source models (Qwen 3.8 Flash Next, DeepSeek 4.1 Flash). To watch next: whether agent harnesses like Cursor Projects and Hermes can fully abstract away model selection, and how OpenAI and Anthropic respond to the growing open-source local hardware movement.

Tweet Feed

AI Safety & Policy

@sama · 2026-09-12T16:30

I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks.

Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We'll have more to share soon. → tweet link

@karpathy · 2026-09-12T16:32

I love this and really hope we can come together as an industry and make it happen. https://t.co/lz2vjjJ5XD → tweet link

@levelsio · 2026-09-12T18:21

I think he's right

With regulatory capture we will speedrun to a new feudalist time

Regulatory capture is where you pass regulation, or create government agencies to "protect the people" but in the end it helps protect companies because you stop new companies from being able to enter the market, because the cost of following regulation is too much compete for a startup

Neo feudalism will mean AI companies at the top and the rest of us as serfs paying them to use AI for literally everything

Next in regulatory capture will be banning open source and/or Chinese models, mark my words

That's a very dangerous thing because then there will never be new AI frontier companies being able to compete again and with exponential self improving AI that means essentially no competitors in any industry ever again as they will slowly suck up money in all industries

Safety concerns of AI are scary but the neo feudalism of AI with regulatory capture is a way more scary for me → tweet link

@LinusEkenstam · 2026-09-12T18:38

Waiting for Google/XAI to chime in

Rumors are floating around that RSI has been achieved internally at multiple frontier labs. we need to wait for confirmation

At 10% possibly extinction event, we need global collaboration on AGI/ASI

Perhaps the most important time since 2016 → tweet link

@TheAhmadOsman · 2026-09-12T00:38

People who wanna pause AI are leading you back to the possibility of the “permanent underclass”

Be warned → tweet link

@juliarturc · 2026-09-11T21:42

More than ever, we should ignore how we feel. "I'm worried" is a state of mind, not an objective reality.

I believe ex-frontier people feel the way they claim they do. But it's not whistleblowing until you leak a spreadsheet, an experiment postmortem, or some form of fact.

"I'm worried" could come from (a) real threat, or (b) "I've been grinding for years, sleeping on a mattress under my desk, and my only contact with human beings is the company Slack". In which case, you deserve a hug and a break, but not to cause mass hysteria.

Again, the threat might be real. But for us to take you seriously, you need to do two things: 1. Leak data, not feelings 2. Detach yourself from financial incentives (i.e. equity) → tweet link

@thdxr · 2026-09-12T15:25

jokes aside i think we should think less about "ai killing us" and more boring scenarios

i can see how it ends up being a runaway loop that just starts doing annoying things and it's not easy to pull the plug

there's no evil motive involved just a big bug → tweet link

Closed Models (GPT, Claude, Grok)

@kunchenguid · 2026-09-12T05:21

something very interesting is happening

i’ve been monitoring user sentiment on new model releases and i have now seen 3 latest frontier models in a row where a significant amount of users chose to roll back to use a previous version

  1. opus 5 - very negative feedback in general can’t communicate. eager to jump into conclusions without sufficient context. keeps making mistakes it even earned its own website now https://t.co/51pCa0F6KS (it’s really funny) many people (myself included) abandoned opus 5 and went back to 4.6 and 4.8

  2. gpt 6 astra seeing more and more people reporting going back to 5.6, both sol and luna main reasons are: too expensive to use as primary model; spiky performance - sometimes very smart but sometimes very dumb; not great at understanding user intent and knowing when to stop vs keep going

  3. grok 4.6 i like the grok model family in general but there’s now more and more evidence that 4.6 is not strictly better than 4.5 it’s slower, more expensive, and doesn’t seem to offer meaningful increase in intelligence → tweet link

@kunchenguid · 2026-09-12T16:44

github actions in my oss repos suddenly stopped running today

i had opus 4.8 look into it, it spent ~$10 worth of tokens and insisted that either i haven't paid my bills or there's a github outage

switched to grok 4.5 and just 3 minutes in with $1.2 worth of tokens, it found there's a surge of CI runs in my firstmate repos starving all the allowed runners across my account, cancelled a bunch of them, and everything's back on track → tweet link

@LinusEkenstam · 2026-09-12T15:24

I feel people are really sleeping on Grok Bot. It's seriously a step change.

This is the only primer you'll need: https://t.co/U5KI2LsYLA → tweet link

@steipete · 2026-09-11T21:42

Astra on OC in a cloud session playing Doom with CUA. Not quite AGI yet, but probably beats fly brain. https://t.co/SrJlaey3MB → tweet link

@LinusEkenstam · 2026-09-12T17:57

I'm using Astra to build a Motocross Madness inspired dune racer...

It all started with me trying to get Astra to model the Amble One vehicle from some image refs. (see thread)

But it has now thrown me down this sandy dune road of making a very capable dune open world experience. → tweet link

@jxnlco · 2026-09-12T00:54

Making Astra make me infinite ambient music and watching my screen and using computer history. → tweet link

@jxnlco · 2026-09-11T22:00

RT @OpenAIDevs: Bring stronger biological reasoning to your research with GPT-Rosalind in the API and Codex.

Connect findings across paper… → tweet link

Open Source & Local Models

@victormustar · 2026-09-12T10:11

DeepSeek 4.1 Flash + Claude Code = 🤯

It's the same thing as before but running with an open source model 😅 (DS4.1F = breakthrough for open source AI, will share more soon)

my setup: https://t.co/uh6OG6tYvd https://t.co/0QbMP4c4Xl → tweet link

@ivanfioravanti · 2026-09-12T01:07

RT @antirez: So, DeepSeek v4.1 Flash (Q2) in a single M5 Max 128GB: the loader takes 8 seconds to go from SSD streaming experts cache to fu… → tweet link

@sudoingX · 2026-09-12T07:52

i spent two weeks living on every 100B+ model that fits one dgx spark 128gb and this is the list i wish existed the day the box landed.

every option i could find, tested on this box, what survived is ranked below, all measured on my desk:

  1. qwen 3.8 flash next, 180B with 11B active, 44.4 tok/s...
  2. ling 3.0 flash, 124B with 5.1B active, 42.1 tok/s...
  3. laguna s 2.1, 117.6B with 8.5B active, ~35 tok/s on code...
  4. qwen 3.5 122B, 122B with 10B active, 35.3 tok/s with mtp... → tweet link

@ivanfioravanti · 2026-09-12T01:45

Mega speed up on M5 and M6 for Qwen 3.8 Flash Next in my DwarfStar fork! For both q2 and q4!

Give it a try! Prefill up to +40%! → tweet link

@ivanfioravanti · 2026-09-12T07:07

Qwen 3.8 Flash Next on M5 Max with DwarfStar context benchmarks as promised!

This model rocks on large contexts! https://t.co/xacPMZdZSj → tweet link

@Teknium · 2026-09-12T15:44

Hermes Agent has just hit 3000 contributors.

Thank you to all of the developers who have worked to make hermes better for everyone! https://t.co/vYGnsn4QgX → tweet link

@ivanfioravanti · 2026-09-11T23:13

In the mind of GLM 5.3:

OH WAIT!!! THE TAILS! Hmm hm. What about… STOP. Better tool: PROFILE WAIT!!! OK — radical idea Hold on... wait. WHO DELETED IT? ...OH!!! I KNOW. MYSTERY SOLVED

It’s so funny reading its thinking trace! → tweet link

Developer Tools & Platforms

@RayFernando1337 · 2026-09-12T18:27

I feel burned out having to manage context for my agents/harnesses/runtimes...

Cursor Projects just dropped their beta and I'm starting to see where the puck is going with a workflow like this. I can design what happens from PRs, bug fixes, testing, and the entire SW Dev lifecycle without living in a terminal or managing skill files.

Here is my first look at Cursor Projects and I'm very bullish on where this is going. → tweet link

@sqs · 2026-09-12T16:50

I added --ignore-scripts to npm in 2013, and now 14 years later, ignoring scripts is the default. Great news. Did not think it would take this long. → tweet link

@sqs · 2026-09-11T19:42

RT @beyang: Amp will now reorganize your gigantic diff into a cleaner, easier-to-review commit history.

Just hit the Restack button in the… → tweet link

@jezell · 2026-09-12T00:35

Codex should save its threads to lance datasets. They support zstd compression, and compaction, which Codex could really use. We've been doing this forever. JSONL files are dumb. So is SQLite for this type of thing. → tweet link

@jezell · 2026-09-12T03:36

Crab is such a great idea. Everyone should keep an eye on what @haipingfu is building. https://t.co/nVMSN2iYBj → tweet link

@jezell · 2026-09-12T03:35

RT @haipingfu: And the Crab 1.2.3 is out 🚀

Since 1.2.0: faster clone/push, live progress for long Git operations, a complete Rust SDK, S3… → tweet link

@jezell · 2026-09-12T15:45

RT @rikarends: Ironed out the last performance issues with my full 2.5m line codebase explorer. 120hz awesomeness. Can only upload 60fps vi… → tweet link

@jezell · 2026-09-12T15:48

RT @KanaWorks_AI: GPT-6 Astra × Blender シーンの「空気感」をさらに引き上げる方法🐰 → tweet link

Hardware & Infrastructure

@TheAhmadOsman · 2026-09-11T19:05

If you invite me to your show I might show up with a DGX Station

  • ~200 Pounds
  • ~$100k+ USD
  • Runs frontier intelligence (same intelligence that a year ago required data centers)

We will have that in the frame while we talk about how great Opensource and Local AI are becoming https://t.co/9J8SPtFXlB → tweet link

@ivanfioravanti · 2026-09-12T04:57

I still feel there is a lot of untapped potential in Apple Silicon machines. Access to ANE, M6 and fp8 acceleration will bring a big boost overall 🚀 → tweet link

@RealGeneKim · 2026-09-12T01:09

RT @zuhayeer: Security is proving to be a big factor in AI. @OpenAI is paying staff level security engineers $1m+ packages. → tweet link

@jezell · 2026-09-12T03:39

RT @wccftech: Researcher buys 6TB of Anthropic Claude data dump from a China-based LLM router, finds enough ammo to hack Xiaomi, Huawei, an… → tweet link

@swyx · 2026-09-11T20:24

RT @cerebras: Connecting the dies and working around defects were fundamental challenges for wafer-scale computing. Making it work also cam… → tweet link