e/acc | Harkonnen Slop Machine

Joined December 2019
Sama’s right - revenue matters. But AGI’s brand must stay clean. OpenAI = AGI, not TikTok. That focus is whats keeping ChatGPT popular despite Google pulling ahead a little.
i get the vibe here, but... we do mostly need the capital for build AI that can do science, and for sure we are focused on AGI with almost all of our research effort. it is also nice to show people cool new tech/products along the way, make them smile, and hopefully make some money given all that compute need. when we launched chatgpt there was a lot of "who needs this and where is AGI". reality is nuanced when it comes to optimal trajectories for a company.
26
14,368
Hard pass to any company that ends up using this for anything whatsoever. This is dystopian stuff. Also the glitch on its face (0:25-0:35 seconds, which will get better perhaps with time but hard to forget), just makes it extremely creepy.
The whole point of building Tavus has been simple: talking to a machine should feel as natural as talking to a friend or coworker. It’s hard to describe all the tiny nuances that make a conversation feel human. The little expressions. Moving around in your chair. Knowing when to speak and when to listen. The dance of it all. Griffin is by far the closest anyone has come to a model that can capture those nuances. The first time I saw it being used, I had no idea I was watching our model rather than just a normal video call. I’m so incredibly proud of this team and what they’ve built.
1
131
Very cool. Great first step towards solving interpretability.
New Anthropic research: Natural Language Autoencoders. Models like Claude talk in words but think in numbers. The numbers—called activations—encode Claude’s thoughts, but not in a language we can read. Here, we train Claude to translate its activations into human-readable text.
135
1. Be an Oracle employee 2. Find out you have been laid off, go back to the office to return your laptop. 3. Get bombed by IRGC. As if things weren’t bad enough already for tech employees.
JUST IN - Iran says it will attack major U.S. tech companies across the Middle East, including Apple, Microsoft, Google, Meta, Nvidia, Oracle, Tesla, HP, Intel and IBM — The Hill
178
Everyone ATP has their own favourite model, and the favourites switch based on internet vibes. Nobody has the time to do a Random Forest thing on LLMs in their day-to-day, who’d use this?
New in M365 Copilot: Council. You can run multiple models on the same prompt at the same time, so you can see where they align and diverge, and understand what each adds.
64
I genuinely want to know which problem this AI engineer was working on for three nights straight at a startup built to solve compliance.
when you step into the office and your founding AI engineer is on his 3rd all nighter this team never stops shipping
107
Lead them to paradise.
Sam Altman has a pretty simple succession plan for OpenAI: Hand off the company to an AI model. If the goal is for artificial intelligence to become so advanced that it can run companies, he asks, then why not his own? “I would never stand in the way of that,” he says. “I should be the most willing to do that.” Read more from our conversation with Sam Altman: forbes.com/sites/richardniev… (Photo: Cody Pickens for Forbes) #Forbes250
75
Easy to say this when they literally force everyone on instagram to “accidentally” click some post from threads. No one can say they know someone who actively uses threads. Just like all of other products of Zuck, this is totally fabricated.
Twitter is dying and threads is replacing it.
63
Two things that might have flipped this from ChatGPT to Gemini: 1. Apple cares the most about privacy, Google must have agreed to hand over control on input and output. 2. Risk of Siri getting consumed by ChatGPT, Google must have agreed to let Apple take the lead on this.
The deal has been finalized: Apple will integrate Gemini into iOS. "After careful evaluation, we determined that Google's technology provides the most capable foundation for Apple Foundation Models and we're excited about the innovative new experiences it will unlock for our users," Apple wrote.
81
Taxing unrealised gains as liquid money is insane economics.
California’s wealth tax is indeed ridiculous. Take Ilya Sutskever as an example. He has $20B+ worth of OpenAI equity now. Let’s say he owns 1/3 of SSI, worth $10B+. He has to give California 5% x (20+10) = $1.5B. And he cannot pay with OpenAI/SSI equity. He has to pay in cash! SSI has (only) raised $3B. This is how California drives founders away.
1
79
Moving forward, Agents would be the primary consumers of every interface that exists today - apps, websites, basically the entire internet. Humans would interact with a much more abstracted version of everything, inputting not more than a prompt of text anywhere.
AI agents will be a big part of how we shop in the not-so-distant future. To help lay the groundwork, we partnered with Shopify, Etsy, Wayfair, Target and Walmart to create the Universal Commerce Protocol, a new open standard for agents and systems to talk to each other across every step of the shopping journey. And coming soon, UCP will power native checkout so you can buy directly on AI Mode and the @Geminiapp.
76
GPT-5.2 Pro solves Erdős Problem #3 (open since 1948). Accepted in discussion by Terence Tao. These problems resist templates and brute force. If this holds, it’s a real shift in mathematical discovery.
Weekend win: The proof I submitted for Erdos Problem #397 was accepted by Terence Tao. The proof was generated by GPT 5.2 Pro and formalized with Harmonic. Many open problems are sitting there, waiting for someone to prompt ChatGPT to solve them:
2
231
Explains a lot
Amazon, $AMZN, AWS CEO says replacing young employees with AI is "one of the dumbest ideas," and bad for business: "At some point, the whole thing explodes on itself."
213
Extremely concerning rhetoric. One can have differences on immigration policy, but actively calling out for violence against a particular group is pure racial hatred. @IndianEmbassyUS @DrSJaishankar should take it up with @USAndIndia and @FBI.
This guy took to the internet to tell the world he wanted to deport Indians from the United States because others may attack them. First of all, why would I want to keep those who would violently attack others in my country? There's no logic to this but hatred.
1
111
I was having an SFW health-oriented conversation with Gemini Flash, and for no-reason at all it redacted the response after generating. The guardrails on Gemini Flash make the experience so much worse than it has to be. Whatever safety classifiers they are using absolutely suck.
1
127
Trump’s 100k H1B fee upheld by Federal Judge, an Obama appointee. Expect more immigration crackdown after this. news.bloomberglaw.com/daily-…
1
206
Pretty cool thing by OpenAI. People love getting an analytical take on them, and this makes ChatGPT feel like a personal assistant. Also, if this is what they are putting out publicly, wonder how much personality mapping they have done internally.
Your Year with ChatGPT! Now rolling out to everyone in the US, UK, Canada, New Zealand, and Australia who have reference saved memory and reference chat history turned on. Just make sure your app is updated.
120
Pretraining of GPT-6 & Gemini 4 Pro has begun. Residents need to cooperate.
BREAKING - San Francisco hit by huge electricity outage, affecting 130,000 residents: power provider Aerial view of San Francisco blackout (📸 Tyler Brown) READ: bit.ly/49oeqnu
104
Waymo discovered an edge case today, the citywide power outage shut down essential infrastructure (especially traffic lights) that its autonomous system depends on for navigation. Eventually waymo suspended all operations in SF to avoid any mishaps.
The power outage in San Francisco has caused Waymo’s self-driving cars to come to a halt, leading to traffic disruptions across the city.
166
Keenly waiting for GPT-5.2, GPT-5.2-Codex and Gemini 3 Pro. But as of now, Claude Opus 4.5 leads the way.
We estimate that, on our tasks, Claude Opus 4.5 has a 50%-time horizon of around 4 hrs 49 mins (95% confidence interval of 1 hr 49 mins to 20 hrs 25 mins). While we're still working through evaluations for other recent models, this is our highest published time horizon to date.
219