Max Buckley

Is Claude flirting with me?

Everytime I claude --dangerously-skip-permissions I think of this: youtube.com/watch?v=b6_IZK…

Nicolay Gerold is the Juggernaut of AI, a regular human empowered by the deity Cyttorak, becoming a literal human juggernaut. He possesses superhuman strength and durability, and is virtually immune to most physical attacks. In his day job he is an entrepreneur, AI engineer and podcaster based in Germany. It was a pleasure hosting Nicolay on the Signal and Stories podcast alongside my cohost @BorisMeinardus. We talked startups in Germany, OpenClaw, coding agents, and much more. Check it out! New Podcast Episode: youtube.com/watch?v=xkNx-G…

Exa Zurich is now hiring ML Research Engineers and Systems Engineers! Have you got a very particular set of skills? Are you an outstanding ML mage, or a systems engineering paladin looking to join a party to take on harder content and level up? We're building the search engine for the AI age — a neural search engine with its own independent index, purpose-built for AI agents. We're growing fast, the problems are hard, and the team is cracked. Open to living in Zurich, Switzerland? Bonus points if your skill tree is specced into: machine learning, search, natural language processing, and/or low-level GPU/CUDA optimisation. ⚠️ Warning: these are extremely high agency roles. You will be expected and empowered to take action from your very first day. History waits for no one! Apply now: jobs.ashbyhq.com/exa?locationId…

We booked a whole IMAX theater in SF for the premiere of "Project Hail Mary". We have room for 250 people. Join to see how to save the world from impending doom, link below 👇

Excited that several Google researchers have launched Exa's new Zurich research office! They'll focus on exploring wild new web-scale retrieval methods. If you want to redesign how search works with mountains in the background: exa․ai/careers

Spitting out amazing pull requests all over the place and repo owners be like:

"The more I practice, the luckier I seem to get" — or the art of engineering your career for maximum luck (of the Irish ☘️). The first episode of Boris Meinardus’s and my new podcast is now live on YouTube! youtu.be/sjz04zTTAPI

In 2016, two organizations spent roughly the same amount: ~$150–160M on marketing for Batman v Superman ~$100–150M in annual operating costs at DeepMind One produced opening-weekend revenue and short-lived buzz. The other produced (amongst other things): • 200M global viewers  • Worldwide frontpage news • National AI policy responses • A Tribeca documentary • A Nobel Prize downstream • A decade of compounding brand equity Yet only one is treated as “marketing”. I argue that many shortsighted financial analyses treat R&D purely as a cost to be minimized, completely failing to recognize that it can be a more powerful marketing and recruitment tool than dedicated marketing and HR programmes. Marketing theory has no practical framework to measure it — no means to quantify the brand equity, earned media, or recruitment value generated by a research breakthrough the way we'd measure a campaign. Essay here: maxbuckley.ai/blog/rnd-marke…

AMLD 2026: "Information Retrieval and Knowledge Management for the AI Age" Workshop Quick reminder to get your tickets for AMLD 2026 if you haven't already! Next week, Applied Machine Learning Days (AMLD) will be taking place at EPFL in Lausanne, Switzerland. AMLD is Switzerland's premier AI conference. I will be teaching a workshop on Thursday alongside fellow Googler Felix Kroner, Google DeepMind's Jovan Andonov and Philippe Schlattner, and NVIDIA's Mickael Ide. The workshop, "Information Retrieval and Knowledge Management for the AI Age," will explore how representation learning, when combined with vector search, enables discovery across any data modality. Making vector search scale to large datasets while maintaining sub-second latency requires special attention; as such, we will take you through a range of techniques and technologies to help you achieve that goal—including approximate nearest neighbor (ANN) algorithms, quantization, embedding model research, and GPU acceleration. This will enable you to build the next generation of multimodal search engines, recommender systems, and RAG systems. 2026.appliedmldays.org Hope to see you there!

Claude Code + GSD implemented a research project I'd been putting off for 6 months. In days with occasional feedback and pointers from me. Not hype — quiet a step change compared to my previous vibe coding experience. If you haven't tried metaprompting frameworks with Claude Code, you should check them out. github.com/glittercowboy/…

The transformer paper is your ML rifle. You should be able to fieldstrip it blindfolded, under pressure, half-asleep. New blog post on nailing attention for interviews 🫡 @maxwbuckley/is-attention-all-you-need-to-prepare-for-ml-interviews-830742f6d2ba?postPublishedType=initial" target="_blank" rel="nofollow noopener">medium.com/@maxwbuckley/i…

8 hours → 87 min → 13 min New post on optimizing a NeurIPS 2024 Concept Bottleneck Model repo. Turns out the bottleneck wasn't the GPU — it was disk I/O, Python loops, and graph breaks. medium.com/p/making-stoch…

He travelled from Tokyo to Zurich to make it happen! Buckle up, @BorisMeinardus fans, we have a taste of things to come. Me and Boris, who you may already know from YouTube, are in the process of setting up a new podcast YouTube channel "Signal and Stories". The channel will feature the two of us and a range of guests from across the industry discussing all things AI and careers. The channel is up on YouTube, but doesn't have any content yet. You can subscribe now to get notified when the first videos drop. I will also share them on here. Get in on the ground floor! @signalandstories" target="_blank" rel="nofollow noopener">youtube.com/@signalandstor… Who would you like to see as a guest? Drop your suggestions in the comments!

@aurielws Brother from another mother! 😆

@MaxWBuckley Welcome my brother 🤓

Does continued pre-training require rerunning your whole post-training stack? Spoilers: looks like no, can leverage the magic of task vectors! arxiv.org/abs/2410.10739

Great book to read if you find yourself a little confused about the state of the world Often described as one of the best books on international relations. There are many very interesting lectures by the author, John Mearsheimer, to be found online. Can also be found on Audible

@natolambert I didn‘t realise you were previously an Australian politician 🤣 grokipedia.com/page/Nathan_La… When entity reconciliation goes wrong…

Fun thing to do. Search for yourself on Grokipedia. If you don't have a page it will prompt you to "Request this article". The vanity pages forbidden by Wikipedia are about to go mainstream 😂 grokipedia.com

Why does the Apple Watch love to decide I am rowing? When walking, it says ‘looks like you are walking, would you like to record a walk’. But every few days it just starts recording a row….

@YuriKushch Better late than never 😅

@MaxWBuckley Finally, here as well 👋