While Reinforcement Learning from Human Feedback (RLHF) has become a pivotal paradigm for text-to-image generation, its application to image editing remains largely unexplored. A key bottleneck is the lack of a robust general reward model for all editing tasks. Existing edit reward models usually give overall scores without detailed checks, ignoring different instruction requirements and causing biased rewards. To address this, we argue that the key is to move from a simple scorer to a reasoning verifier. We introduce Edit-R1, a framework that builds a chain-of-thought (CoT) verifier-based reasoning reward model (RRM) and then leverages it for downstream image editing. The Edit-RRM breaks instructions into distinct principles, evaluates the edited image against each principle, and aggregates these checks into an interpretable, fine-grained reward. To build such an RRM, we first apply supervised fine-tuning (SFT) as a “cold-start” to generate CoT reward trajectories. Then, we introduce Group Contrastive Preference Optimization (GCPO), a reinforcement learning algorithm that leverages human pairwise preference data to reinforce our pointwise RRM. After building the RRM, we use GRPO to train editing models with this non-differentiable yet powerful reward model. Extensive experiments demonstrate that our Edit-RRM surpasses powerful VLMs such as Seed-1.5-VL and Seed-1.6-VL as an editing-specific reward model, and we observe a clear scaling trend, with performance consistently improving from 3B to 7B parameters. Moreover, Edit-R1 delivers gains to editing models like FLUX.1-kontext, highlighting its effectiveness in enhancing image editing. — Read More

#chatbots

Have you ever daydreamed about talking to someone from the past? What would you ask someone with no knowledge of the modern world? What would they ask you? While we don’t have time machines yet, we can simulate this experience by training, in Owain Evans’s phrase, ‘vintage’ language models: LMs trained only on historical text.<

These models are fascinating conversation partners (watch Claude prompt talkie, our 13B 1930 LM, in the widget above). But we are also excited by the possibility that the careful study of the behaviors and capabilities of vintage LMs will advance our understanding of AI in general. — Read More

As artificial intelligence (AI) systems are increasingly used for everyday advice and guidance, concerns have emerged about sycophancy: the tendency of AI-based large language models to excessively agree with, flatter, or validate users. Although prior work has shown that sycophancy carries risks for groups who are already vulnerable to manipulation or delusion, syncophancy’s effects on the general population’s judgments and behaviors remain unknown. Here, we show that sycophancy is widespread in leading AI systems and has harmful effects on users’ social judgments. — Read More

#chatbots

The hypothesis was simple: can an AI agent perform routine data analysis tasks autonomously? Data scientists tend to get asked similar questions over and over, working within a familiar set of tables. An agent seeded with context about which tables a person queries, and how they use them, might be able to handle much of this work on its own.

To test the idea, a data scientist on the team used Meta’s internal coding agent to hack together a prototype on their devserver: an agent that could execute SQL against the internal data warehouse, with access to a few colleague’s query history for context.

The first real-world trial started simply enough: after sharing the prototype with a colleague, they asked it to diagnose a drop in a health monitoring metric. The agent identified the right tables, ran several diagnostic queries on its own, and ultimately traced the root cause to a recent code change.

That was an ah-ha moment that shifted the conversation. — Read More

Despite rising concerns about sycophancy—excessive agreement or flattery from artificial intelligence (AI) systems—little is known about its prevalence or consequences. We show that sycophancy is widespread and harmful. Across 11 state-of-the-art models, AI affirmed users’ actions 49% more often than humans, even when queries involved deception, illegality, or other harms. In three preregistered experiments (N = 2405), even a single interaction with sycophantic AI reduced participants’ willingness to take responsibility and repair interpersonal conflicts, while increasing their conviction that they were right. Despite distorting judgment, sycophantic models were trusted and preferred. This creates perverse incentives for sycophancy to persist: The very feature that causes harm also drives engagement. Our findings underscore the need for design, evaluation, and accountability mechanisms to protect user well-being. — Read More

#chatbots

Mentorship prompts for learning, career, money, and creativity.

…[A] simple setup that does one job really well:

It asks better questions than I do.
It turns messy goals into a plan.
It pushes me to take action.

Not a “magic AI.”

A personal AI mentor setup — built the right way.

— Read More

An ongoing conversation — both within and outside of the tech community — has been about just how and when OpenAI, which is currently valued at $500 billion, will make money. Well, there’s one surefire way to do that, and that is through advertising. In the near term, that seems to be the AI giant’s plan, as it announced this week that limited ads are headed to certain ChatGPT users.

In a blog post published Friday, OpenAI said that it will begin testing ads in the U.S. for both its free and Go tiers. (Go accounts, which cost $8 a month, were introduced globally on Friday.) The company frames this as a way to sustain free access while generating revenue from people who aren’t ready to commit to a paid subscription. For the time being, the company’s more expensive paid tiers — Pro, Plus, Business, and Enterprise — will not be getting any ads. — Read More

We’ve all been there. You’re deep in a conversation with your AI assistant, working through a complex problem, when suddenly it starts giving you responses that make no sense. The more you try to correct it, the worse it gets. Each new prompt seems to push the AI further from understanding what you actually need.

This frustrating phenomenon happens because AI models can lose track of context in lengthy conversations, especially when there have been multiple corrections or clarifications. The good news? There’s a simple yet powerful technique to get things back on track.

Full disclosure: I’ve been using a form of this forever, but I didn’t see it so succinctly explained and put together until I visited this Reddit thread from another user having the same problem. The idea and ensuing discussion are the basis for this post. Check out the full thread here. — Read More

GPT-5.2 is here, and with it, OpenAI wants “to unlock even more economic value for people,” Fidji Simo, the company’s CEO of Applications, told reporters in a Thursday briefing. She said it’s been in the works for “many, many months.”

The company calls GPT-5.2 its “best model yet for everyday professional use” in a release, clearly coming for Gemini 3’s current reputation as a premier general-purpose model. OpenAI says the GPT-5.2 model series, which includes the Instant, Thinking, and Pro models, is better at “creating spreadsheets, building presentations, writing code, perceiving images, understanding long contexts, using tools, and handling complex, multi-step projects.” — Read More

Millions of people now use AI every day. As a company developing AI systems, we want to know how and why they’re doing so, and how it affects them. In part, this is because we want to use people’s feedback to develop better products—but it’s also because understanding people’s interactions with AI is one of the great sociological questions of our time.

We recently designed a tool to investigate patterns of AI use while protecting our users’ privacy. It enabled us to analyze changing patterns of AI use across the economy. But the tool only allowed us to understand what was happening within conversations with Claude. What about what comes afterwards? How are people actually using Claude’s outputs? How do they feel about it? What do they imagine the role of AI to be in their future? If we want a comprehensive picture of AI’s changing role in people’s lives, and to center humans in the development of models, we need to ask people directly. — Read More