dotAI 2024
2024-10-17..2024-10-18
Paris, France
dotAI ·
[dotAI 2017, dotAI 2018, dotAI 2024, dotAI 2025, dotAI 2026]
who it's for
machine learning engineers, mlops engineers, research scientists, data scientists, infrastructure and platform engineers, product managers, heads of AI and CTOs
topics
multimodal language models; real-time voice agents; open-source multi-agent systems; serverless inferencing infrastructure; model quantization and edge deployment; vector databases for training; prompt engineering and output controllers; training foundation image models; data table vectorization; fairness via latent decorrelation; autoscaling and scale-to-zero; proof-search and RL pipelines
| title | author | length |
|---|---|---|
| Building with OpenAI: What's Ahead | Romain Huet, Katia Gil Guzman | 23m |
Demo shows a real-time API that opens a websocket two-way stateful connection to do native speech-to-speech, multimodal I/O and function/tool calls. A live app… | ||
| ZML: ML framework for Zig | Steeve Morin | 13m |
ZML compiles Zig model code via MLIR/XLA and Bazel into production inference artifacts that run on NVIDIA, AMD (ROCm) and TPU, with a rebuilt Torch loader and… | ||
| Skrub: Less data wrangling, more machine learning | Gael Varoquaux | 19m |
Claims that Skrub slashes the 80% data‑prep burden by providing a Python library that converts messy tables into model‑ready features, shipping a table… | ||
| LLMs reasoning and agentic capabilities over time | Stanislas Polu | 19m |
Claim: scaled autoregressive LLMs plus a search loop and verified objective produce surprising leaps — an autoformalize + search-loop pipeline over formal… | ||
| The future of Edge Agents | Pierre Stock | 20m |
Presents running Mistral 3B (3 billion parameters) efficiently on mobile/edge devices by mixed quantization to int8/int5 to fit in a few GB of RAM while… | ||
| Multimodal language models | Neil Zeghidour | 20m |
Kyutai presents an open-source weights and code release for a compact multimodal LLM that combines a neural audio codec, a hierarchical generator and… | ||
| Reality is not an end-to-end prediction problem | Ines Montani | 18m |
Argues that prompts alone fail for production systems and shows a concrete pipeline: use spaCy llm and Prodigy for LLM-assisted annotation, transfer-learn… | ||
| Can AI design a fairer future? | Marcin Detyniecki | 13m |
Uses the COMPAS recidivism example and an insurance pricing use case to show a concrete adversarial technique: an encoder H with HGR penalty that decorrelates… | ||
| Training a foundation image model and deploying it to millions of users | Eliot Andres | 20m |
A custom diffusion model is trained from scratch with a bespoke dataset and architecture, running on 256 H100s and consuming energy comparable to 600 French… | ||
| Democratizing AI with Open Source Multi-Agent Systems | Sri Satish Ambati | 21m |
Claims nightly Gaia benchmarks on an enterprise platform with Claude leading but open‑source models closing the gap, releases of 0.5B 2B 4B models trained on… | ||
| Gain full control of your apps with Open-Source AI | Merve Noyan | 25m |
Claim: smaller open-source models often outcompete larger proprietary ones when distilled, fine-tuned and served with modern tooling. Evidence: Zier 7B hits… | ||
| Serverless inferencing: an infrastructure point of view | Yann Léger | 19m |
Demonstrates a live deploy of Lama 3.1 on A100 with autoscaling from one to three instances and a benchmark that triggers new VMs every 15 seconds, claims real… | ||
| More empathy and health data protection in AI | Laure Seugé, Arthur Talpaert | 15m |
Doctolib announces the Consultation Assistant launch: a pipeline of signal processing, ASR, information extraction, medical summarization and coding that… | ||
| Improving open language models at a practical size | Armand Joulin | 18m |
Gemma 2 ships as 2B, 9B and 27B models and claims top-in-class size-normalized performance by training small models against teacher probability distributions… | ||
| Getting LLMs to do what you want: Output controllers | Meryem Arik | 13m |
A deterministic method called JSON DeRuLO enforces a JSON schema at inference time by applying token masking and a cheap character‑level filter so the model… | ||
| AI is your personal co-founder | Marjolaine Grondin | 16m |
Chronicles a concrete workflow: a failed app pivoted into building an AI assistant, then automations that actually run—Zapier co-pilot for invoices, WhatsApp… | ||
| Is Mojo programming language just a hype? | Maxim Zaks | 11m |
Mojo claims to be a practical successor for heavy compute workloads rather than mere hype: born at Modular because teams were tired of hand-writing MLIR… | ||
| AI deployment strategies: balancing efficiency and environmental impact | Ori Pekelman | 16m |
Claims that LLM apps currently emit about 0.03 megatons a day (~11 megatons/year) and that adding 2 million H100s sold this year and Grace Hoppers drives… | ||
| Multi-agent collaboration in practice: make AI code an app for you | Grigorij Dudnik | 09m |
Claims an open source framework called Clean Coder that actually writes and updates app code is usable in production and wrote most of the startup code for… | ||
| Rethinking AI: Building What Matters, Not Just What's Possible | Audrey Roy Greenfeld | 09m |
A working parody-generation pipeline uses LLMs with function/tool calling, a Python class for inputs/outputs, and dual agents—a writer and editor—to generate… | ||
| Crossing the multimodal chasm | Daniel Madalitso Phiri | 08m |
Argues that crossing the multimodal chasm requires composing smaller models into a plumbing pipeline rather than one monolithic foundation model, and shows… | ||
| The RAod to intent-driven automation | Johannes Dienst | 08m |
Demonstrates a working JARVIS‑style system that automates UIs by translating intents into mouse/keyboard actions and evaluative checks, successfully generating… | ||
| Message from the Secretary of State for AI and Digital Affairs | Clara Chappaz | 11m |
France reports 2.5 billion euros invested since 2018 and counts over 1,000 AI startups in Paris, claims major labs and VCs have relocated here, and plans an AI… | ||
| nothing matches | ||