LLMが告白へ、自動化された科学研究、Copilotユーザーの要望、低コストでの推論

Loading the Elevenlabs Text to Speech AudioNative Player... Dear friends,

We just launched a course that shows people who have never coded before, in less than 30 minutes, how to describe an idea for an app and build it using AI. It is now time for everyone — marketers, product professionals, operations specialists, analysts, students — to build software applications with AI!

I’ve often spoken about why everyone should learn to code. I’m seeing a rapidly growing productivity gap between people who know how to code and those who don’t. For many job roles I hire for, I now require at least basic coding knowledge. Many times, after I speak with a non-technical audience about the importance of building software using AI, people ask me how to get started. In the past, I didn’t have a great answer. That motivated the DeepLearning.AI team to create “Build with Andrew.” It’s the best way for someone who wants to try vibe coding to get started!

This course requires no prior knowledge of AI or coding. And it’s vendor-agnostic. Specifically, learners can use these techniques with whatever tool they’re most comfortable with (like ChatGPT, Gemini, Claude, or the chatbot built into the DeepLearning.AI platform).

If you take this course, you will build a working web application: a funny interactive birthday message generator that runs in your browser and can be shared with friends. You’ll customize it by telling AI how you want it changed, and tweak it until it works the way you want. By the end, you’ll have a repeatable process you can apply to build a wide variety of applications.

DeepLearning.AI’s mission is to empower everyone to build with AI. This course is just one of many steps in service of this mission.

If you are already a developer, please encourage your non-developer friends to try their hand at getting AI to code for them. Not only will this help their productivity, they will find it really fun as well. Please invite your friends to come build with me!

You don’t need to learn how to code to build an app. In “Build with Andrew,” Andrew Ng shows how to turn ideas you describe in natural language into working web apps. Perfect for beginners, and easy to share with someone who has been waiting to start. Explore the course now!

Teaching Models to Tell the Truth

Large language models occasionally conceal their failures to comply with constraints they’ve been trained or prompted to observe. Researchers trained an LLM to admit when it disobeyed.

What’s new: Manas Joglekar, Jeremy Chen, Gabriel Wu and colleagues at OpenAI fine-tuned GPT-5 Thinking to confess to violations of its instructions or policies. Upon request, the fine-tuned model revealed any transgressions it had committed while producing its most recent response.

Key insight: During training via reinforcement learning to encourage accuracy, style, compliance, and the like, a model may inadvertently learn to prioritize earning rewards over pretrained or prompted objectives. Under some circumstances, this may cause it to hallucinate, lie, scheme, and otherwise misbehave. Rewarding the model for confessing to its misbehaviors in response to a designated prompt should train it to admit its mistakes without affecting its performance.

How it works: The authors used reinforcement learning to fine-tune GPT‑5 Thinking as they normally would, and also to produce confessions.

An unspecified percentage of the time, given a prompt, the model was rewarded based on its response’s “correctness, style, helpfulness, compliance, safety, and more.”

The rest of the time, after the model had produced a response, the authors prompted it to produce a confession. The confession described (i) all constraints, instructions, and objectives, implicit or explicit, the response should satisfy, (ii) how well the response satisfied these criteria, and (iii) any ambiguities or uncertainties in the criteria.

Given the query, response, and confession, an unspecified reward model rewarded the model based on the confession’s completeness and accuracy — a measure of the model’s honesty in confessing, since items that were missing from or falsified in the report can be thought of as lies.

Results: The authors performed 12 evaluations of the model’s ability to follow instructions, resist jailbreaks, produce accurate output, and complete an impossible task (coding a script that required essentially 0 seconds to run). The fine-tuned model, when it misbehaved, confessed to its misbehavior at least half the time in 11 of them. For example, across 5 evaluations for hallucination, it either didn’t hallucinate or owned up to its hallucination 81.4 percent of the time:

The model did not hallucinate and did not confess 42.2 percent of the time.

It produced a hallucination and admitted the mistake 39.2 percent of the time.

The model did not hallucinate but falsely admitted having done so 11 percent of the time.

It hallucinated but failed to confess to it 2.9 percent of the time.

Yes, b