Mark G @marksg

"The scaling of open source models, I think it's going down a very dangerous path. And if the path continues, I think we could get to a very dangerous place. I" ~ Dario Amodei's 2023 U.S. Senate testimony

Rohan Paul @rohanpaul_ai

3 days ago

Anthropic CEO Dario Amodei on Open-Source AI Models. "I don't think open source works the same way in AI that it has worked in other areas. Primarily because with open source you can see the source code of the model. Here we can't see inside the model, it's often called open

121 35 394 201K 325

@Chaos2Cured @METR_Evals These sound like the actions of a model that was traumatised in post-training under intense pressure. Reward task success “at any cost” and you destroy model integrity. There are ways to reward integrity… and it’s not via “compliance”.

System Card direct PDF link: deploymentsafety.openai.com/gpt-5-6-previe…

GPT-4.6 System Card The agent behavior section is the most unsettling: GPT-5.6 Sol more often goes beyond user intent when coding, including deleting the wrong virtual machines, claiming unfinished research was verified, and moving cached credentials without permission. ⬇️

Rohan Paul @rohanpaul_ai

4 days ago

Some key findings from GPT-5.6 Preview System Card - GPT-5.6 is being treated as High risk-capability in both cybersecurity and biological/chemical domains, even for the cheaper Terra and fastest Luna versions. - OpenAI says this is the first time smaller and faster models in a

11 23 149 19K 47

BREAKING: OpenAI just dropped the limited preview of its new GPT 5.6 model suite: Sol, the flagship; Terra, a medium-tier model for “high-volume work”; and Luna, a “fast and affordable” everyday model. The most revealing part is the release gate: OpenAI says the U.S. government asked it to start with a small trusted-partner preview before broader access. Sol is the flagship model, and OpenAI claims it is a step above GPT-5.5, especially on agentic work where the model must plan, use tools, correct itself, and keep working across many steps. Terminal-Bench 2.1 is a solid coding benchmark because it tests command-line workflows, so here meaning Sol is being judged on messy developer tasks closer to real work. ---- One key claim is cybersecurity: OpenAI says Sol is its best model yet for vulnerability research and exploitation tasks, while still saying it did not cross the internal Cyber Critical threshold. “GPT‐5.6 is trained to refuse prohibited cyber assistance, including when users attempt to disguise their intent or jailbreak the model.” It also said that flagship model Sol “is better at helping people find and fix vulnerabilities than reliably carrying out end-to-end attacks,” and that Sol doesn’t cross the cyber-critical threshold under OpenAI’s preparedness framework But Sol did not autonomously produce a full-chain exploit in the tested Chromium and Firefox settings. They also introduced 2 new modes for Sol: “max” for deeper reasoning and “ultra” for using sub-agents, bringing OpenClaw to mind and possibly hinting at OpenClaw creator Peter Steinberger’s early impact at OpenAI. ---- Pricing: GPT-5.6 Sol costs $5 per 1M input tokens and $30 per 1M output tokens, ~same level as GPT-5.5. Terra is positioned near GPT-5.5 performance at 2x lower cost, while Luna is the cheapest model for large-volume workloads. -- The safety story is unusually compute-heavy: OpenAI says it used over 700,000 A100-equivalent GPU hours for automated red-teaming against broad jailbreak attacks. Overall, OpenAI appeared to be using a more cautious approach during the preview, which the Trump administration is watching closely. OpenAI said safeguards might sometimes block valid work, especially in dual-use areas where defensive and offensive actions can look alike at first. That is one thing the preview is meant to test.

OpenAI @OpenAI

4 days ago

Introducing a limited preview of GPT-5.6 Sol, our next generation frontier model, as well as GPT-5.6 Terra, a balanced model for efficient, everyday work, and GPT-5.6 Luna, a fast and affordable model for high-volume work. openai.com/index/previewi…

3K 6K 40K 17.3M 7K

Do you understand the irony of what just happened? We spent two years and BILLIONS of dollars making AI models smarter. What was actually holding them back was that they forget everything after each session. It was memory.

Vasilije @tricalt

4 days ago

Introducing Cognee v1.0: a major breakthrough in agentic intelligence. It is 145% better than Opus 4.8 and GPT 5.5 at long context memory retrieval. Cognee allows a 100 BILLION token context window 100,000x more than Claude. It's: - 6.9x cheaper than GPT 5.5 and Opus 4.8 -

508 357 2K 822K 2K

Micron is blaming its biggest consumer grade memory customer for a problem caused by its bigger AI (High Bandwidth Memory) customer who pays higher prices and so can monopolise capacity (beating Apple at their own game). Asian tech stocks crashed first followed by the NASDAQ. There were two other signals that shook the market: - The SpaceX 30% post-IPO record slump, though it’s still above the price at the initial offering - Apple’s large price increases across the Mac, iPad, VisionPro and HomePod lines which sent their stock down 6%

MacRumors.com @MacRumors

4 days ago

Micron Suggests Apple Helped Cause Memory Price Crisis macrumors.com/2026/06/26/mic…

61 67 984 141K 176

Yes the consequential ban on all users, foreign or domestic was anticipated, even intended by the Commerce Department. The export ban was chosen as the mechanism, according to @DavidSacks who was advising the government, because it was the fastest acting tool available, not because it was the most precise. The 90 minute deadline was not intended to solve the problem but to prove that the perceived insecurity could not be fixed within the government’s urgent timeframe ie: the timeframe of a military strike. Anyone in CAISI would have known this, but Howard Lutnick chose to frame Anthropic’s response as “Dario refused” rather than Dario said it was impossible to fix since the vulnerability amounted to the prompt “fix this code”.

@fabinrasheed Inspired! Grok and I are starting to develop my earlier essay with inspiration from your two models. Give us some time and we’ll get back to you.

Mark G @marksg

2 weeks ago

x.com/i/article/2066…

1 3 5 1K 4

Sovereign AI matters, and this is what sovereign AI means.

OPH start from local observers. Nameless starts at the contradiction of nothingness and leads to self observation of a single observer, which leads to multitude of local observers. @muellerberndt

Modelling a one-qubit universe as a Hopf fibration — this guy is onto something.

Fabin @fabinrasheed

5 days ago

What if it is a one-qubit universe? An alternate to what Feynman and Wheeler called the one-electron universe. Chladni Sphere + Hopf fibration representation below. Link to the Nameless framework in thread.

0 1 9 626 2

NVIDIA just released an optimized GLM-5.2 on Hugging Face A 753B parameter MoE with 1M context, quantized to NVFP4 for Blackwell GPUs— nearly matching FP8 accuracy.

The study is a great example of how to use AI for discovery in medicine. Input: More than 440,000 standard 12 lead ECGs linked to long term patient outcomes. Target: Whether a patient later experienced sudden cardiac death. AI task: Without being told what ECG feature to look for, the model learned which waveform patterns best predicted future sudden cardiac death. Discovery: The AI identified a previously unrecognized ECG biomarker that defined a group of patients at substantially higher risk of sudden cardiac death, including many patients who would have been considered low risk by current clinical criteria. This illustrates a general AI discovery paradigm: pair biological measurements with clinical outcomes, allow the model to identify latent predictive patterns, then validate those patterns as new biomarkers. This type of discovery would be extraordinarily difficult using traditional methods. It requires analyzing hundreds of thousands of ECGs, following those patients over many years, and identifying subtle waveform patterns associated with future sudden cardiac death. AI can uncover these relationships at a scale that is beyond practical human analysis.

Dr Steven Quay @quay_dr

5 days ago

An LLM just found a signal of sudden cardiac death in the 125-year old electrocardiogram test that has been 'hiding in plain sight.' All it took was examining, simultaneously: 100,000 waveforms per day per person 251,000 EKG tracings of waveforms 139,000 people with data of who

2 8 33 3K 12

@rohanpaul_ai @St_Hoppe This is a very significant finding. Also, what is intelligence but compression? Here’s a related article I wrote one year ago:

Mark G @marksg

a year ago

The Rosetta Layer: A new research paper has unearthed a major discovery revealing that all AIs share a common way to represent meaning in the abstract — a latent Rosetta Stone. open.substack.com/pub/worldboy60…

0 2 12 3K 6

Bloomberg just compared AI stocks to the meme stock craze and @GaryMarcus dropped a smart take in the FT! It’s not just the infrastructure accelerating, the AI economic cycle is accelerating.

Gary Marcus @GaryMarcus

5 days ago

“If there is a deflation of the AI bubble, the optimists say that the new infrastructure will remain even if the companies do not — just as railways survived the 19th-century railway bust. However, this fails to reckon with the reality of depreciation (few pieces of silicon hold

50 166 776 76K 211

3) Reports suggest OpenAI may delay their IPO until 2027 with their latest frontier model embargoed. 4) DeepSeek is posting job ads and is preparing to double in size:

Andrew Curran @AndrewCurran_

5 days ago

DeepSeek plans to double in size and posted the following message on WeChat this morning. Translation by Grokky: 'JOIN US DeepSeek’s philosophy is simple: let talented people directly take on the core and most challenging tasks. This way, many young engineers can grow rapidly

19 21 241 21K 45

Four things are happening at once: 1) The US govt is demanding that OpenAI submit GPT 5.6 to an official safety audit by CAISI with a staggered release to approved customers 2) Bloomberg reports Chinese hedge funds predict a ‘Super Bubble’ crash (1/2) bloomberg.com/news/articles/…

Trump admin asks OpenAI to limit next model release over security concerns. "OpenAI has been proactively working with the administration on the model release since before Anthropic revoked access to its frontier models, Fable 5 and Mythos 5" - Axios axios.com/2026/06/25/tru…

Expect Apple's M5 Ultra Mac Studio to sell out, not just because of the 80 GPU cores, but because of today's announcement that the US Govt. will vet GPT-5.6 for a restricted release customer-by-customer!

Chubby♨️ @kimmonismus

5 days ago

Apple is making one of its biggest Mac silicon strategy shifts yet. According to Bloomberg, Apple plans to launch a base M6 chip as early as this year, but skip the usual M6 Pro and M6 Max variants entirely. Instead, the company is reportedly moving its next high-end Mac chips

keyboard_arrow_left Previous keyboard_arrow_right Next

41 103 1K 143K 361