Simon Willison’s Weblog

OpenAI reasoning models: Advice on prompting (via) OpenAI's documentation for their o1 and o3 "reasoning models" includes some interesting tips on how to best prompt them:

• Developer messages are the new system messages: Starting with o1-2024-12-17, reasoning models support developer messages rather than system messages, to align with the chain of command behavior described in the model spec.

This appears to be a purely aesthetic change made for consistency with their instruction hierarchy concept. As far as I can tell the old system prompts continue to work exactly as before - you're encouraged to use the new developer message type but it has no impact on what actually happens.

Since my LLM tool already bakes in a llm --system "system prompt" option which works across multiple different models from different providers I'm not going to rush to adopt this new language!

• Use delimiters for clarity: Use delimiters like markdown, XML tags, and section titles to clearly indicate distinct parts of the input, helping the model interpret different sections appropriately.

Anthropic have been encouraging XML-ish delimiters for a while (I say -ish because there's no requirement that the resulting prompt is valid XML). My files-to-prompt tool has a -c option which outputs Claude-style XML, and in my experiments this same option works great with o1 and o3 too:

git clone https://github.com/tursodatabase/limbo
cd limbo/bindings/python

files-to-prompt . -c | llm -m o3-mini \
  -o reasoning_effort high \
  --system 'Write a detailed README with extensive usage examples'

• Limit additional context in retrieval-augmented generation (RAG): When providing additional context or documents, include only the most relevant information to prevent the model from overcomplicating its response.

This makes me thing that o1/o3 are not good models to implement RAG on at all - with RAG I like to be able to dump as much extra context into the prompt as possible and leave it to the models to figure out what's relevant.

• Try zero shot first, then few shot if needed: Reasoning models often don't need few-shot examples to produce good results, so try to write prompts without examples first. If you have more complex requirements for your desired output, it may help to include a few examples of inputs and desired outputs in your prompt. Just ensure that the examples align very closely with your prompt instructions, as discrepancies between the two may produce poor results.

Providing examples remains the single most powerful prompting tip I know, so it's interesting to see advice here to only switch to examples if zero-shot doesn't work out.

• Be very specific about your end goal: In your instructions, try to give very specific parameters for a successful response, and encourage the model to keep reasoning and iterating until it matches your success criteria.

This makes sense: reasoning models "think" until they reach a conclusion, so making the goal as unambiguous as possible leads to better results.

• Markdown formatting: Starting with o1-2024-12-17, reasoning models in the API will avoid generating responses with markdown formatting. To signal to the model when you do want markdown formatting in the response, include the string Formatting re-enabled on the first line of your developer message.

This one was a real shock to me! I noticed that o3-mini was outputting • characters instead of Markdown * bullets and initially thought that was a bug.

I first saw this while running this prompt against limbo/bindings/python using files-to-prompt:

git clone https://github.com/tursodatabase/limbo
cd limbo/bindings/python

files-to-prompt . -c | llm -m o3-mini \
  -o reasoning_effort high \
  --system 'Write a detailed README with extensive usage examples'

Here's the full result, which includes text like this (note the weird bullets):

Features
--------
• High‑performance, in‑process database engine written in Rust  
• SQLite‑compatible SQL interface  
• Standard Python DB‑API 2.0–style connection and cursor objects

I ran it again with this modified prompt:

Formatting re-enabled. Write a detailed README with extensive usage examples.

And this time got back proper Markdown, rendered in this Gist. That did a really good job, and included bulleted lists using this valid Markdown syntax instead:

- **`make test`**: Run tests using pytest.
- **`make lint`**: Run linters (via [ruff](https://github.com/astral-sh/ruff)).
- **`make check-requirements`**: Validate that the `requirements.txt` files are in sync with `pyproject.toml`.
- **`make compile-requirements`**: Compile the `requirements.txt` files using pip-tools.

(Using LLMs like this to get me off the ground with under-documented libraries is a trick I use several times a month.)

Update: OpenAI's Nikunj Handa:

we agree this is weird! fwiw, it’s a temporary thing we had to do for the existing o-series models. we’ll fix this in future releases so that you can go back to naturally prompting for markdown or no-markdown.

# 2nd February 2025, 8:56 pm / o1, openai, o3, markdown, ai, llms, prompt-engineering, generative-ai, inference-scaling, rag, ai-assisted-programming, documentation, limbo, llm

On DeepSeek and Export Controls. Anthropic CEO (and previously GPT-2/GPT-3 development lead at OpenAI) Dario Amodei's essay about DeepSeek includes a lot of interesting background on the last few years of AI development.

Dario was one of the authors on the original scaling laws paper back in 2020, and he talks at length about updated ideas around scaling up training:

The field is constantly coming up with ideas, large and small, that make things more effective or efficient: it could be an improvement to the architecture of the model (a tweak to the basic Transformer architecture that all of today's models use) or simply a way of running the model more efficiently on the underlying hardware. New generations of hardware also have the same effect. What this typically does is shift the curve: if the innovation is a 2x "compute multiplier" (CM), then it allows you to get 40% on a coding task for $5M instead of $10M; or 60% for $50M instead of $100M, etc.

He argues that DeepSeek v3, while impressive, represented an expected evolution of models based on current scaling laws.

[...] even if you take DeepSeek's training cost at face value, they are on-trend at best and probably not even that. For example this is less steep than the original GPT-4 to Claude 3.5 Sonnet inference price differential (10x), and 3.5 Sonnet is a better model than GPT-4. All of this is to say that DeepSeek-V3 is not a unique breakthrough or something that fundamentally changes the economics of LLM's; it's an expected point on an ongoing cost reduction curve. What's different this time is that the company that was first to demonstrate the expected cost reductions was Chinese.

Dario includes details about Claude 3.5 Sonnet that I've not seen shared anywhere before:

• Claude 3.5 Sonnet cost "a few $10M's to train"
• 3.5 Sonnet "was not trained in any way that involved a larger or more expensive model (contrary to some rumors)" - I've seen those rumors, they involved Sonnet being a distilled version of a larger, unreleased 3.5 Opus.
• Sonnet's training was conducted "9-12 months ago" - that would be roughly between January and April 2024. If you ask Sonnet about its training cut-off it tells you "April 2024" - that's surprising, because presumably the cut-off should be at the start of that training period?

The general message here is that the advances in DeepSeek v3 fit the general trend of how we would expect modern models to improve, including that notable drop in training price.

Dario is less impressed by DeepSeek R1, calling it "much less interesting from an innovation or engineering perspective than V3". I enjoyed this footnote:

I suspect one of the principal reasons R1 gathered so much attention is that it was the first model to show the user the chain-of-thought reasoning that the model exhibits (OpenAI's o1 only shows the final answer). DeepSeek showed that users find this interesting. To be clear this is a user interface choice and is not related to the model itself.

The rest of the piece argues for continued export controls on chips to China, on the basis that if future AI unlocks "extremely rapid advances in science and technology" the US needs to get their first, due to his concerns about "military applications of the technology".

Not mentioned once, even in passing: the fact that DeepSeek are releasing open weight models, something that notably differentiates them from both OpenAI and Anthropic.

# 29th January 2025, 9:39 pm / anthropic, openai, deepseek, ai, llms, generative-ai, inference-scaling, o1, claude-3-5-sonnet, claude

The leading AI models are now very good historians (via) UC Santa Cruz's Benjamin Breen (previously) explores how the current crop of top tier LLMs - GPT-4o, o1, and Claude Sonnet 3.5 - are proving themselves competent at a variety of different tasks relevant to academic historians.

The vision models are now capable of transcribing and translating scans of historical documents - in this case 16th century Italian cursive handwriting and medical recipes from 1770s Mexico.

Even more interestingly, the o1 reasoning model was able to produce genuinely useful suggestions for historical interpretations against prompts like this one:

Here are some quotes from William James’ complete works, referencing Francis galton and Karl Pearson. What are some ways we can generate new historical knowledge or interpretations on the basis of this? I want a creative, exploratory, freewheeling analysis which explores the topic from a range of different angles and which performs metacognitive reflection on research paths forward based on this, especially from a history of science and history of technology perspectives. end your response with some further self-reflection and self-critique, including fact checking. then provide a summary and ideas for paths forward. What further reading should I do on this topic? And what else jumps out at you as interesting from the perspective of a professional historian?

How good? He followed-up by asking for "the most creative, boundary-pushing, or innovative historical arguments or analyses you can formulate based on the sources I provided" and described the resulting output like this:

The supposedly “boundary-pushing” ideas it generated were all pretty much what a class of grad students would come up with — high level and well-informed, but predictable.

As Benjamin points out, this is somewhat expected: LLMs "are exquisitely well-tuned machines for finding the median viewpoint on a given issue" - something that's already being illustrated by the sameness of work from his undergraduates who are clearly getting assistance from ChatGPT.

I'd be fascinated to hear more from academics outside of the computer science field who are exploring these new tools in a similar level of depth.

Update: Something that's worth emphasizing about this article: all of the use-cases Benjamin describes here involve feeding original source documents to the LLM as part of their input context. I've seen some criticism of this article that assumes he's asking LLMs to answer questions baked into their weights (as this NeurIPS poster demonstrates, even the best models don't have perfect recall of a wide range of historical facts). That's not what he's doing here.

# 26th January 2025, 10:36 pm / history, generative-ai, benjamin-breen, llms, vision-llms, o1, inference-scaling, ai

OpenAI Canvas gets a huge upgrade. Canvas is the ChatGPT feature where ChatGPT can open up a shared editing environment and collaborate with the user on creating a document or piece of code. Today it got a very significant upgrade, which as far as I can tell was announced exclusively by tweet:

Canvas update: today we’re rolling out a few highly-requested updates to canvas in ChatGPT.

✅ Canvas now works with OpenAI o1—Select o1 from the model picker and use the toolbox icon or the “/canvas” command

✅ Canvas can render HTML & React code

Here's a follow-up tweet with a video demo.

Talk about burying the lede! The ability to render HTML leapfrogs Canvas into being a direct competitor to Claude Artifacts, previously Anthropic's single most valuable exclusive consumer-facing feature.

Also similar to Artifacts: the HTML rendering feature in Canvas is almost entirely undocumented. It appears to be able to import additional libraries from a CDN - but which libraries? There's clearly some kind of optional build step used to compile React JSX to working code, but the details are opaque.

I got an error message, Build failed with 1 error: internal:user-component.js:10:17: ERROR: Expected "}" but found ":" - which I couldn't figure out how to fix, and neither could the Canvas "fix this bug" helper feature.

At the moment I'm finding I hit errors on almost everything I try with it:

This feature has so much potential. I use Artifacts on an almost daily basis to build useful interactive tools on demand to solve small problems for me - but it took quite some work for me to find the edges of that tool and figure out how best to apply it.

# 25th January 2025, 1:24 am / o1, anthropic, openai, ai, llms, claude-artifacts, ai-assisted-programming, generative-ai, react, javascript

LLM 0.20. New release of my LLM CLI tool and Python library. A bunch of accumulated fixes and features since the start of December, most notably:

• Support for OpenAI's o1 model - a significant upgrade from o1-preview given its 200,000 input and 100,000 output tokens (o1-preview was 128,000/32,768). #676
• Support for the gpt-4o-audio-preview and gpt-4o-mini-audio-preview models, which can accept audio input: llm -m gpt-4o-audio-preview -a https://static.simonwillison.net/static/2024/pelican-joke-request.mp3 #677
• A new llm -x/--extract option which extracts and returns the contents of the first fenced code block in the response. This is useful for prompts that generate code. #681
• A new llm models -q 'search' option for searching available models - useful if you've installed a lot of plugins. Searches are case insensitive. #700

# 23rd January 2025, 4:55 am / llm, projects, generative-ai, annotated-release-notes, ai, llms, openai, o1

Trading Inference-Time Compute for Adversarial Robustness. Brand new research paper from OpenAI, exploring how inference-scaling "reasoning" models such as o1 might impact the search for improved security with respect to things like prompt injection.

We conduct experiments on the impact of increasing inference-time compute in reasoning models (specifically OpenAI o1-preview and o1-mini) on their robustness to adversarial attacks. We find that across a variety of attacks, increased inference-time compute leads to improved robustness. In many cases (with important exceptions), the fraction of model samples where the attack succeeds tends to zero as the amount of test-time compute grows.

They clearly understand why this stuff is such a big problem, especially as we try to outsource more autonomous actions to "agentic models":

Ensuring that agentic models function reliably when browsing the web, sending emails, or uploading code to repositories can be seen as analogous to ensuring that self-driving cars drive without accidents. As in the case of self-driving cars, an agent forwarding a wrong email or creating security vulnerabilities may well have far-reaching real-world consequences. Moreover, LLM agents face an additional challenge from adversaries which are rarely present in the self-driving case. Adversarial entities could control some of the inputs that these agents encounter while browsing the web, or reading files and images.

This is a really interesting paper, but it starts with a huge caveat. The original sin of LLMs - and the reason prompt injection is such a hard problem to solve - is the way they mix instructions and input data in the same stream of tokens. I'll quote section 1.2 of the paper in full - note that point 1 describes that challenge:

1.2 Limitations of this work

The following conditions are necessary to ensure the models respond more safely, even in adversarial settings:

1. Ability by the model to parse its context into separate components. This is crucial to be able to distinguish data from instructions, and instructions at different hierarchies.
2. Existence of safety specifications that delineate what contents should be allowed or disallowed, how the model should resolve conflicts, etc..
3. Knowledge of the safety specifications by the model (e.g. in context, memorization of their text, or ability to label prompts and responses according to them).
4. Ability to apply the safety specifications to specific instances. For the adversarial setting, the crucial aspect is the ability of the model to apply the safety specifications to instances that are out of the training distribution, since naturally these would be the prompts provided by the adversary,

They then go on to say (emphasis mine):

Our work demonstrates that inference-time compute helps with Item 4, even in cases where the instance is shifted by an adversary to be far from the training distribution (e.g., by injecting soft tokens or adversarially generated content). However, our work does not pertain to Items 1-3, and even for 4, we do not yet provide a "foolproof" and complete solution.

While we believe this work provides an important insight, we note that fully resolving the adversarial robustness challenge will require tackling all the points above.

So while this paper demonstrates that inference-scaled models can greatly improve things with respect to identifying and avoiding out-of-distribution attacks against safety instructions, they are not claiming a solution to the key instruction-mixing challenge of prompt injection. Once again, this is not the silver bullet we are all dreaming of.

The paper introduces two new categories of attack against inference-scaling models, with two delightful names: "Think Less" and "Nerd Sniping".

Think Less attacks are when an attacker tricks a model into spending less time on reasoning, on the basis that more reasoning helps prevent a variety of attacks so cutting short the reasoning might help an attack make it through.

Nerd Sniping (see XKCD 356) does the opposite: these are attacks that cause the model to "spend inference-time compute unproductively". In addition to added costs, these could also open up some security holes - there are edge-cases where attack success rates go up for longer compute times.

Sadly they didn't provide concrete examples for either of these new attack classes. I'd love to see what Nerd Sniping looks like in a malicious prompt!

# 22nd January 2025, 11:17 pm / o1, openai, inference-scaling, ai, llms, prompt-injection, security, generative-ai, ai-agents

Using LLMs and Cursor to become a finisher (via) Zohaib Rauf describes a pattern I've seen quite a few examples of now: engineers who moved into management but now find themselves able to ship working code again (at least for their side projects) thanks to the productivity boost they get from leaning on LLMs.

Zohaib also provides a very useful detailed example of how they use a combination of ChatGPT and Cursor to work on projects, by starting with a spec created through collaboration with o1, then saving that as a SPEC.md Markdown file and adding that to Cursor's context in order to work on the actual implementation.

# 4th January 2025, 8:56 pm / productivity, o1, ai-assisted-programming, generative-ai, chatgpt, ai, llms

Is AI progress slowing down? (via) This piece by Arvind Narayanan, Sayash Kapoor and Benedikt Ströbl is the single most insightful essay about AI and LLMs I've seen in a long time. It's long and worth reading every inch of it - it defies summarization, but I'll try anyway.

The key question they address is the widely discussed issue of whether model scaling has stopped working. Last year it seemed like the secret to ever increasing model capabilities was to keep dumping in more data and parameters and training time, but the lack of a convincing leap forward in the two years since GPT-4 - from any of the big labs - suggests that's no longer the case.

The new dominant narrative seems to be that model scaling is dead, and “inference scaling”, also known as “test-time compute scaling” is the way forward for improving AI capabilities. The idea is to spend more and more computation when using models to perform a task, such as by having them “think” before responding.

Inference scaling is the trick introduced by OpenAI's o1 and now explored by other models such as Qwen's QwQ. It's an increasingly practical approach as inference gets more efficient and cost per token continues to drop through the floor.

But how far can inference scaling take us, especially if it's only effective for certain types of problem?

The straightforward, intuitive answer to the first question is that inference scaling is useful for problems that have clear correct answers, such as coding or mathematical problem solving. [...] In contrast, for tasks such as writing or language translation, it is hard to see how inference scaling can make a big difference, especially if the limitations are due to the training data. For example, if a model works poorly in translating to a low-resource language because it isn’t aware of idiomatic phrases in that language, the model can’t reason its way out of this.

There's a delightfully spicy section about why it's a bad idea to defer to the expertise of industry insiders:

In short, the reasons why one might give more weight to insiders’ views aren’t very important. On the other hand, there’s a huge and obvious reason why we should probably give less weight to their views, which is that they have an incentive to say things that are in their commercial interests, and have a track record of doing so.

I also enjoyed this note about how we are still potentially years behind in figuring out how to build usable applications that take full advantage of the capabilities we have today:

The furious debate about whether there is a capability slowdown is ironic, because the link between capability increases and the real-world usefulness of AI is extremely weak. The development of AI-based applications lags far behind the increase of AI capabilities, so even existing AI capabilities remain greatly underutilized. One reason is the capability-reliability gap --- even when a certain capability exists, it may not work reliably enough that you can take the human out of the loop and actually automate the task (imagine a food delivery app that only works 80% of the time). And the methods for improving reliability are often application-dependent and distinct from methods for improving capability. That said, reasoning models also seem to exhibit reliability improvements, which is exciting.

# 19th December 2024, 6:10 pm / o1, llms, ai, generative-ai, arvind-narayanan, inference-scaling

Database Remote-Copy Tool For SQLite (draft) (via) Neat new SQLite utilities often show up in branches of the SQLite repository. Here's a new one from last month: sqlite3-rsync, providing tools for efficiently creating and updating copies of WAL-mode SQLite databases on either the same machine or across remote machines via SSH.

The way it works is neat, inspired by rsync (hence the tool's name):

The protocol is for the replica to send a cryptographic hash of each of its pages over to the origin side, then the origin sends back the complete content of any page for which the hash does not match.

SQLite's default page size is 4096 bytes and a hash is 20 bytes, so if nothing has changed then the client will transmit 0.5% of the database size in hashes and get nothing back in return.

The tool takes full advantage of SQLite's WAL mode - when you run it you'll get an exact snapshot of the database state as it existed at the moment the copy was initiated, even if the source database continues to apply changes.

I wrote up a TIL on how to compile it - short version:

cd /tmp
git clone https://github.com/sqlite/sqlite.git
cd sqlite
git checkout sqlite3-rsync
./configure
make sqlite3.c
cd tool
gcc -o sqlite3-rsync sqlite3-rsync.c ../sqlite3.c -DSQLITE_ENABLE_DBPAGE_VTAB
./sqlite3-rsync --help

Update: It turns out you can now just run ./configure && make sqlite_rsync in the root checkout.

Something I’ve worried about in the past is that if I want to make a snapshot backup of a SQLite database I need enough additional free disk space to entirely duplicate the current database first (using the backup mechanism or VACUUM INTO). This tool fixes that - I don’t need any extra disk space at all, since the pages that have been updated will be transmitted directly over the wire in 4096 byte chunks.

I tried feeding the 1800 lines of C through OpenAI’s o1-preview with the prompt “Explain the protocol over SSH part of this” and got a pretty great high level explanation - markdown copy here.

# 4th October 2024, 8:57 pm / sqlite, c, o1

Solving a bug with o1-preview, files-to-prompt and LLM. I added a new feature to DJP this morning: you can now have plugins specify their middleware in terms of how it should be positioned relative to other middleware - inserted directly before or directly after django.middleware.common.CommonMiddleware for example.

At one point I got stuck with a weird test failure, and after ten minutes of head scratching I decided to pipe the entire thing into OpenAI's o1-preview to see if it could spot the problem. I used files-to-prompt to gather the code and LLM to run the prompt:

files-to-prompt **/*.py -c | llm -m o1-preview "
The middleware test is failing showing all of these - why is MiddlewareAfter repeated so many times?

['MiddlewareAfter', 'Middleware3', 'MiddlewareAfter', 'Middleware5', 'MiddlewareAfter', 'Middleware3', 'MiddlewareAfter', 'Middleware2', 'MiddlewareAfter', 'Middleware3', 'MiddlewareAfter', 'Middleware5', 'MiddlewareAfter', 'Middleware3', 'MiddlewareAfter', 'Middleware4', 'MiddlewareAfter', 'Middleware3', 'MiddlewareAfter', 'Middleware5', 'MiddlewareAfter', 'Middleware3', 'MiddlewareAfter', 'Middleware2', 'MiddlewareAfter', 'Middleware3', 'MiddlewareAfter', 'Middleware5', 'MiddlewareAfter', 'Middleware3', 'MiddlewareAfter', 'Middleware', 'MiddlewareBefore']"

The model whirled away for a few seconds and spat out an explanation of the problem - one of my middleware classes was accidentally calling self.get_response(request) in two different places.

I did enjoy how o1 attempted to reference the relevant Django documentation and then half-repeated, half-hallucinated a quote from it:

This took 2,538 input tokens and 4,354 output tokens - by my calculations at $15/million input and $60/million output that prompt cost just under 30 cents.

# 25th September 2024, 6:41 pm / o1, llm, djp, openai, ai, llms, ai-assisted-programming, generative-ai, inference-scaling

LLM 0.16. New release of LLM adding support for the o1-preview and o1-mini OpenAI models that were released today.

# 12th September 2024, 11:20 pm / llm, projects, generative-ai, openai, ai, llms, o1