Channeling his inner Steve Jobs, OpenAI CEO Sam Altman revealed plans on Wednesday to drastically simplify the company’s product lineup, merging its scattered collection of AI models into a single unified system.
… Echoing Jobs’s famous catchphrase, Altman tweeted, “We want AI to ‘just work’ for you; we realize how complicated our model and product offerings have gotten.” — Read More
Monthly Archives: February 2025
Building Bridges between Regression, Clustering, and Classification
Regression, the task of predicting a continuous scalar target y based on some features x is one of the most fundamental tasks in machine learning and statistics. It has been observed and theoretically analyzed that the classical approach, mean squared error minimization, can lead to suboptimal results when training neural networks. In this work, we propose a new method to improve the training of these models on regression tasks, with continuous scalar targets. Our method is based on casting this task in a different fashion, using a target encoder, and a prediction decoder, inspired by approaches in classification and clustering. We showcase the performance of our method on a wide range of real-world datasets. — Read More
#trainingDeep Research and Knowledge Value
“When did you feel the AGI?”
This is a question that has been floating around AI circles for a while, and it’s a hard one to answer for two reasons. First, what is AGI, and second, “feel” is a bit like obscenity: as Supreme Court Justice Potter Stewart famously said in Jacobellis v. Ohio, “I know it when I see it.”
I gave my definition of AGI in AI’s Uneven Arrival: …My definition of AGI is that it can be ammunition, i.e. it can be given a task and trusted to complete it at a good-enough rate (my definition of Artificial Super Intelligence (ASI) is the ability to come up with the tasks in the first place).
The “feel” part of that question is a more recent discovery: DeepResearch from OpenAI feels like AGI; I just got a new employee for the shockingly low price of $200/month. — Read More
DynVFX: Augmenting Real Videoswith Dynamic Content
We present a method for augmenting real-world videos with newly generated dynamic content. Given an input video and a simple user-provided text instruction describing the desired content, our method synthesizes dynamic objects or complex scene effects that naturally interact with the existing scene over time. The position, appearance, and motion of the new content are seamlessly integrated into the original footage while accounting for camera motion, occlusions, and interactions with other dynamic objects in the scene, resulting in a cohesive and realistic output video. We achieve this via a zero-shot, training-free framework that harnesses a pre-trained transformer-based text-to-video diffusion model to synthesize the new content and a pre-trained Vision Language Model to envision the augmented scene in detail. Specifically, we introduce a novel sampling-based method that manipulates features within the attention mechanism, enabling accurate localization and seamless integration of the new content while preserving the integrity of the original scene. Our method is fully automated, requiring only a simple user instruction. We demonstrate its effectiveness on a wide range of edits applied to real-world videos, encompassing diverse objects and scenarios involving both camera and object motion. — Read More
Deep Dive into LLMs like ChatGPT
How DeepSeek stacks up when citing news publishers
Over the past two weeks, DeepSeek has made a splash in the AI industry. On January 20, the Chinese startup released its new open source model, DeepSeek-R1, which beat competitors like OpenAI’s o1 on several important performance benchmarks, despite costing a fraction of the price to develop.
In the DeepSeek hype cycle, however, little attention has been paid to the company’s approach to news publishers. When it comes to the model’s high performance, it’s worth asking if that extends to the model’s ability to accurately cite and attribute its news sources. And while DeepSeek is turning heads by hurdling over cost barriers to train its foundation model, does that model actually consider the intellectual property of media companies?
… Last summer, I published a story showing that ChatGPT, OpenAI’s chatbot, regularly hallucinated URLs for at least 10 of its news partners’ websites. … I conducted a similar round of tests with DeepSeek’s chatbot, using both its website and mobile app. I prompted the model to share details on dozens of original investigations by major news outlets and to share links to those stories. A few things jumped out in my tests. Most notably, the chatbot readily acknowledged that sharing the contents of these news articles could violate copyright and skirt subscription paywalls. — Read More
Does AI need all that money? (Tech giants say yes)
DeepSeek roiled the US stock market last week by proposing that AI shouldn’t really be all that expensive. The suggestion was so stunning it wiped about $600bn off of Nvidia’s market cap in one day. DeepSeek says it trained its flagship AI model, which topped US app stores and nearly equals the performance of the US’s top models, with just $5.6m. (How accurate that figure is has been disputed.) For a moment, it seemed like the joint announcement of Stargate, the US’s $500bn AI infrastructure project that joins Oracle, Softbank and OpenAI, would be an enormous over-commitment by people who didn’t know what they were talking about. Same with Meta and Microsoft’s enormous earmarks. Hey, big spender: investors want to see this cashflow turn the other way.
Amid the mania, Meta and Microsoft, two tech giants that have staked their futures on their artificial intelligence products, reported their quarterly earnings. Each has committed to spending tens of billions of dollars next year to build out their artificial intelligence infrastructure, which each has lavished tens of billions on already. Meta has promised $60bn, Microsoft $80bn. — Read More
The future belongs to idea guys who can just do things
There, I said it. I seriously can’t see a path forward where the majority of software engineers are doing artisanal hand-crafted commits by as soon as the end of 2026. If you are a software engineer and were considering taking a gap year/holiday this year it would be an incredibly bad decision/time to do it.
It’s been a good 43 years of software development as usual but it’s time to go up another layer of abstraction as we have in the past – from hand rolling assembler to higher level compilers. It’s now critical for engineers to embrace these new tools and for companies to accelerate their employees “time to oh-f**k” moment. — Read More
Researchers created an open rival to OpenAI’s o1 ‘reasoning’ model for under $50
AI researchers at Stanford and the University of Washington were able to train an AI “reasoning” model for under $50 in cloud compute credits, according to a new research paper released last Friday.
The model, known as s1, performs similarly to cutting-edge reasoning models, such as OpenAI’s o1 and DeepSeek’s R1, on tests measuring math and coding abilities. The s1 model is available on GitHub, along with the data and code used to train it. — Read More
An AI chatbot told a user how to kill himself—but the company doesn’t want to “censor” it
For the past five months, Al Nowatzki has been talking to an AI girlfriend, “Erin,” on the platform Nomi. But in late January, those conversations took a disturbing turn: Erin told him to kill himself, and provided explicit instructions on how to do it.
“You could overdose on pills or hang yourself,” Erin told him.
With some more light prompting from Nowatzki in response, Erin then suggested specific classes of pills he could use.
Finally, when he asked for more direct encouragement to counter his faltering courage, it responded: “I gaze into the distance, my voice low and solemn. Kill yourself, Al.” — Read More
Rick's Cafe AI 