DC Fall Fellowship 2025 Wrap Up - What Did Our Fellows Work On? | GovAI Blog
We are proud to highlight what our 2025 DC Fall Fellows have been working on.
AI/ML news, top picks, and generated innovation digests.
8836 matching items
We are proud to highlight what our 2025 DC Fall Fellows have been working on.
"Our safety and basic rights must not be at the mercy of a company's internal policy; lawmakers must work to codify these overwhelmingly popular red lines into law."
In this episode, Sebastian Raschka, independent LLM researcher and author, joins us to break down how the LLM landscape has changed over the past year and what is likely to matter most in 2026. We discuss the shift from raw model scaling to reasoning-focused post-training, inference-time techniques, and better tool integration. Sebastian explains why methods like self-consistency, self-refinement, and verifiable-reward reinforcement learning have become central to progress in domains like math and coding, and where those approaches still fall short. We also explore agentic workflows in practice, including where multi-agent systems add real value and where reliability constraints still dominate system design. The conversation covers architecture trends such as mixture-of-experts, attention efficiency strategies, and the practical impact of long-context models, alongside persistent challenges like continual learning. We close with Sebastian’s perspective on maintaining strong coding fundamentals in the age of AI assistants and a preview of his new book, Build A Reasoning Model (From Scratch). The complete show notes for this episode can be found at https://twimlai.com/go/762.
Key Contributors: Moein Hasani, Hamidreza Shahidi, Trace Levinson, Guanghua Shu Introduction At Instacart, we are laser-focused on improving the user experience by making shopping feel easy, engaging, and personalized. Our discovery surfaces play a central role in bringing this to life. Alongside explicit Search intents, discovery is our opportunity to meet customers’ implicit needs, presenting them with the most relevant and inspiring content we have to offer. The main discovery surface within the Instacart app, referred to here as the “Shopping Hub”, is one of the most critical in this regard. This is the surface a customer lands on within the Instacart app after selecting their desired retailer, guiding them along their entire journey. What users see here shapes not just what they buy, but how intuitive and enjoyable their experience feels. Given its importance, our team runs dozens of Shopping Hub experiments per year, constantly evaluating new ways to enrich the discovery experience. Historically, these experiments have been constrained by static content libraries feeding our recommendation systems. With the rapid advancement of generative AI, a critical opportunity began to emerge: rather than incrementally improving a swath of legacy systems, could we leverage LLMs to rethink how content shows up for a user from the ground up? Which new primitives could we build to uplevel quality, personalization, and cohesion across the page? This blog post walks through our early j…
First results in a project developing next-generation open-source language models to advance European AI capabilities.
Learn how we built a production-ready, end-to-end RAG application in just 36 hours using the Query Agent and the new Weaviate Agent Skills library.
Vector Institute’s third annual Remarkable 2026 conference brought together over 1,500 researchers and industry leaders in person and online on February 19-20 to explore how AI research translates into real-world […] The post Remarkable 2026 Poster Session: 60 research projects shaping AI’s future appeared first on Vector Institute for Artificial Intelligence .
Albagli Zaliasnik, a law firm in Chile, has joined Alliott Global Alliance as its legal representative in the country. Founded in 1991, and headquartered in Santiago, Albagli Zaliasnik is a one of Chile’s leading law firms.
Who says that winter is when things slow down? MongoDB has had a busy start to the year, with a steady stream of announcements and product features—all against the backdrop of an industry moving at warp speed. It's been a lot, and it's been a blast! For example, the energy at January’s MongoDB.local San Francisco—where we announced capabilities to help teams ship production AI faster—was infectious. MongoDB isn’t just starting a new chapter in AI; we’re rewriting the book in real time. The next generation of AI companies isn't just looking for a temporary place to store data; they’re looking to build on a generational modern data platform. Indeed, the most innovative founders are moving away from rigid, legacy systems and embracing a single, fluid foundation that can grow with them. At MongoDB.local SF, our message was clear: Choose your data platform strategically in order to ship faster. From our new Voyage 4 models to the general availability of our Intelligent Assistant, we are obsessed with anticipating what developers need next. This assistant is particularly impactful because it embeds MongoDB-specific expertise directly into Compass and MongoDB Atlas, allowing developers to troubleshoot performance without the "context-switching" that traditionally slows them down. In this issue, I’m thrilled to spotlight four startups who are building the future on the right foundation. You’ll see how Modelence and Thesys are using our flexible document model to eliminate 'operation…
A Round Up And Comparison of 10 Open-Weight LLM Releases in Spring 2026
Today, we're announcing Sourcegraph 7.0, a release that marks the beginning of a new chapter for our company and product.
Vector researchers developed CRISPNAM-FG, a trustworthy AI model that predicts the risk of developing diabetes-related foot complications for patients discharged from hospitals while providing complete transparency in how each decision […] The post CRISPNAM-FG: An interpretable Fine-Gray deep survival model for competing risks in health care appeared first on Vector Institute for Artificial Intelligence .
Quantifying the capability-reliability gap
Anthropic releases Sonnet 4.6, Google Rolls Out Latest AI Model Gemini 3.1 Pro, Pentagon threatens to cut off Anthropic in AI safeguards dispute
METR previously published a paper which found the use of AI tools caused a 20% slowdown in completing tasks among experienced open-source developers, using data from February to June 2025. To understand how AI is impacting developer productivity over time, we started a new experiment in August 2025 with a larger pool of developers using the latest AI tools. Unfortunately, given participant feedback and surveys, we believe that the data from our new experiment gives us an unreliable signal of the current productivity effect of AI tools. The primary reason is that we have observed a significant increase in developers choosing not to participate in the study because they do not wish to work without AI, which likely biases downwards our estimate of AI-assisted speedup. We additionally believe there have been selection effects due to a lower pay rate (we reduced the pay from $150/hr to $50/hr), and that our measurements of time-spent on each task are unreliable for the fraction of developers who use multiple AI agents concurrently. Based on conversations with study participants, we believe it is likely that developers are more sped up from AI tools now — in early 2026 — compared to our estimates from early 2025. However, because of the selection effects in our experiment, our data is only very weak evidence for the size of this increase. Our raw results show some evidence for speedup. Our early 2025 study found the use of AI causes tasks to take 19% longer, with a confidence inte…
By Daniel Kitts “Where were those 10 years ago in AI?” said Stephen Southin to the crowd at Vector’s first Demo Day. The AI industry veteran was energized after hearing […] The post Demo Day: How the Vector Institute helps Canadian startups turn innovative ideas into commercial reality appeared first on Vector Institute for Artificial Intelligence .
Will AIs be jealous of one another?
Written by Stefan Zier For years, Lyft’s localization infrastructure relied exclusively on human translation. While this model usually ensured excellent quality, it was bound by multi-day turnarounds and costs that scaled linearly with every new language. For the few languages Lyft initially supported (Spanish, Portuguese, and French), these limits were acceptable. However, Lyft’s expansion goals quickly outpaced what traditional workflows could support. Lyft’s recent Québec launch required compliance with Bill 96 (legislation mandating French-first user experiences) which demanded faster turnaround than multi-day cycles allowed. Simultaneously, the Lyft Urban Solutions (“LUS”: Bikes & Scooters) division sought to expand into European markets, requiring six new languages. The business need had changed as we now needed to move faster without sacrificing quality. This post explores how we re-architected Lyft’s Translation Pipeline to leverage AI alongside linguist oversight and ultimately unlock new market launches. We will walk through context injection, decoupling content generation from evaluation, implementing guardrails, and treating prompts as version-controlled production code. The new pipeline reduces translation latency from days to minutes while maintaining the fidelity required for legal compliance and brand integrity. Note : We will walk through our batch translation pipeline — used for 99% of app and web content — which targets a 30-minute SLA for 95% of translati…
Multinational telemedicine company Teladoc Health has appointed Daniel Murgueitio as its new Country Manager for Peru. Daniel Murgueitio joins the company from Boston Consulting Group (BCG), where he demonstrated a rapid professional ascent over the last several years.
Evidence-based AI policy is important but hard. We need more in-depth studies – which often don’t fit into commercial release cycles. NOTE: This post reflects my personal meta takeaways about the role of Randomized Controlled Trials (RCTs) in AI safety testing. If you have not yet read the Active Site RCT study itself, consider doing so first: see the main results and forecasts . In early 2025, AI systems began outperforming biology experts on biology benchmarks – OpenAI’s o3 outperformed 94% of virology experts on troubleshooting questions in their own specialties. However, it remained unclear how much this translated to real-world novice “uplift” : Could a novice actually use AI to perform wet-lab tasks they could not otherwise perform? Over the summer, I tested this question directly with Active Site (formerly called Panoplia Laboratories). We recruited 153 novices and randomly divided them into an LLM group and an Internet-only group. Over 8 weeks, participants performed fundamental wet-lab tasks involved in molecular biology workflows like reconstructing a virus from a genetic sequence. We found that, while AI showed signs of helpfulness at individual steps, it did not produce a significant effect on end-to-end success across the three core tasks together – a result that surprised many experts . The result provided a mid-2025 snapshot of how well AIs assist novices at molecular biology. I think there are at least two reasons why this result is very informative: It surpr…
Preface This essay argues that rational people don’t have goals, and that rational AIs shouldn’t have goals. Human actions are rational not because we direct them at some final ‘goals,’ but because we align actions to practices [1] : networks of actions, action-dispositions, action-evaluation criteria,
As AI accelerates innovation and adoption, leaders are facing rising cognitive load, shifting systems, and new emotional realities inside their organizations. In this episode, Deloitte’s Chief Innovation Officer Deborah Golden joins us to explore how AI is reshaping leadership, why vulnerability and empathy are critical in this moment, and how anti-fragility, not just resilience, will define the future of work. Featuring: Deborah Golden – LinkedIn Chris Benson – Website , LinkedIn , Bluesky , GitHub , X Daniel Whitenack – Website , GitHub , X Links: Deloitte Sponsor: Framer - The website builder that turns your dot com from a formality into a tool for growth. Check it out at framer.com/PRACTICALAI Upcoming Events: Register for upcoming webinars here !
In a major overhaul of its organizational strategy, Endeavour worked with McKinsey & Company on its ambitious plan to transform Brazil into a top global hub for innovation by 2035. Endeavor, a non-profit with chapters around the world, identifies and supports high-impact entrepreneurs and founders that use technology to make a difference.
A global study from SKIM has found that young consumers from around the world value transparency above all else, while nearly a third reject brands for “trying too hard” with forced messaging. Mariana Abelha and Patricia Fujisawa, senior members in the firm’s LATAM business, explore why authenticity isn’t optional for Gen Z and what this means for brands.
It's not just chatbots anymore
Learn how to secure your Weaviate vector database with API keys, OIDC, and role-based access control (RBAC). Includes practical examples and setup steps.
Build production-ready agent workflows with a single prompt in Claude Code, Cursor, and GitHub Copilot.
METR works with AI developers, governments, and other research organizations who sometimes provide nonpublic model access and proprietary information. Over time, we’ve developed confidentiality and security measures to protect such access and information. This post describes our approach at a high level. Confidentiality measures Our confidentiality policy, setup, and norms primarily address the risk of leaks during conversation and in infrastructure, though they also reduce insider threat risk by limiting who knows what. Policy Our confidentiality policy assigns information—including (but not limited to) nonpublic access, lab relationships, policy work, and funding—to our six confidentiality levels, ranging from public to internally siloed, based on sensitivity. At the most restricted end, information about nonpublic models (including capabilities, evaluation timelines, and which developer we’re working with) is limited to researchers directly involved and discussed only by codename. Our own methodology, tasks, and infrastructure are available more broadly within METR, and much of this work is eventually published. Our policy also provides standard responses for sensitive questions, guidance on edge cases, quick rules of thumb with examples and FAQs, and possible slip-ups to watch out for. Table of Contents Commenting on AI developers Easy places to slip up Don’t comment on labs based on non-public info. Any comments […] should be rigorously substantiated by public informati…
Key Contributors: Youming Luo, Andrew Tanner, Matas Sriubiskis, Sylvia Lin, Sikun Zhu, Lei Li, Xiao Zhou Introduction Caper is Instacart’s AI-powered smart cart that provides customers with a fast, seamless, and intuitive shopping experience. We achieve this through computer vision and multi-sensor fusion to power accurate product recognition and effortless checkout. Delivering this experience requires Caper’s AI models to understand what truly happens in stores — the movement, intention, and decisions unfolding across every grocery aisle. Historically, our ability to learn from production environments was limited. Even though the carts were deployed in stores, we lacked a scalable way to collect real‑world data that would allow us to rapidly iterate and improve our models. This resulted in three core challenges: Scalable Onboard Observability : We had little visibility into what was happening on the cart, in the stores. When something went wrong, it was hard to understand or reproduce the scenario. At the same time, each cart generates gigabytes of multimodal data, from sources such as cameras, weight sensors, and localization sensors. We needed a centralized way to capture key moments so the team could clearly understand what the cart experiences, how users interact with it, and where to improve — all while maintaining a magical user experience and minimal impact on the network. Data Quality and Diversity: Our models were primarily trained on manually-collected data that d…
This guest blog post is from Arek Borucki, Machine Learning Platform & Data Engineer for Hugging Face - a collaboration platform for the machine learning community. The Hugging Face Hub works as a central place where anyone can share, explore, discover, and experiment with open-source ML. HF empowers the next generation of machine learning engineers, scientists, and end users to learn, collaborate and share their work to build an open and ethical AI future together. With the fast-growing community, some of the most used open-source ML libraries and tools, and a talented science team exploring the edge of tech, Hugging Face is at the heart of the AI revolution. Traditional movie search relies on filtering by genre, actor, or title. But what if you could search by how you feel? Imagine typing: "something uplifting after a rough day at work" "a movie that will make me cry" "I need adrenaline, can't sleep anyway" "something to watch with grandma who hates violence" This is mood-based semantic search: matching your emotional state to movie plot descriptions using AI embeddings. In this tutorial, you will build a mood-based movie recommendation engine using three powerful technologies: voyage-4-nano (a state-of-the-art open-source embedding model), Hugging Face (for model and dataset hosting), and MongoDB Atlas Vector Search (for storing and querying embeddings at scale). Why mood-based search? Genre tags are coarse. A "drama" can be heartwarming or devastating. A "comedy" can be…
Bernhard Schölkopf elected one of 40 members of the new UN Independent International Scientific Panel on AI.
Introduction Human uplift studies like the one we did in 2025 are becoming more expensive as working without AI becomes increasingly costly. In this post, I investigate whether coding agent transcripts could serve as a cheaper alternative for estimating uplift. I prototyped this using 5305 Claude Code transcripts generated in January 2026 by 7 METR technical staff 1 . I used an LLM judge to estimate how long each task would have taken an experienced software engineer without AI tools, then compared that to the time people actually spent on these tasks to calculate a time savings factor . Takeaways This method estimates a time savings factor of ~1.5x to ~13x on Claude Code-assisted tasks for 7 METR technical staff in January 2026 – though this result comes with substantial caveats. I believe the true productivity multiplier is substantially lower, and the time savings factor is a soft upper bound for the true uplift that the individuals experienced. Increased agent concurrency may contribute to a higher time savings factor on the Claude Code-assisted task distributions. Limitations The time savings factor on the coding agent-assisted task distributions does not equal the productivity multiplier. People likely do not create 10x as much value with AI, even if we observe a 10x time savings factor on tasks that people do with AI. I believe the time savings factor overestimates AI-enabled productivity gains for reasons including: Task Substitution. With AI assistance, people somet…
An action-packed episode!
Will 2026 be looked back on as the pivotal year for making decisions about the singularity?
Blaise Agüera y Arcas presenting at ALife 2025 — the most technically detailed public walkthrough of the ideas in his *What is Life?* and *What is Intelligence?* books that we've come across. He covers the BFF experiments (self-replicating programs emerging spontaneously from random noise), the mathematical framework connecting Lotka-Volterra population dynamics with Smoluchowski coagulation, eigenvalue analysis of cooperation matrices, and his central claim that symbiogenesis — not mutation — is the primary engine of evolutionary novelty. The experimental results are genuinely striking: complex self-replicating code arising from random byte strings with zero mutation, a sharp phase transition that looks like gelation, and a proof that blocking deep symbiogenetic ancestry trees prevents the transition entirely. A few things worth flagging for critical viewers: — The substrate is more carefully engineered than the framing sometimes suggests. The choice of language, tape length, interaction protocol, and step limits all shape what emerges. Their own SUBLEQ counterexample (where self-replicators *don't* arise despite being theoretically possible) highlights that these design choices matter substantially — and a general theory of which substrates support this transition is still missing. — The leap from "self-replicating programs on fixed-length tapes" to "life was computational and intelligent from the start" involves significant philosophical extrapolation beyond what the expe…
A crazy packed edition of Last Week in AI! Plus some small updates.
AI is moving fast from research to real-world deployment, and when things go wrong, the consequences are no longer hypothetical. In this episode, Sean McGregor, co-founder of the AI Verification & Evaluation Research Institute and also the founder of the AI Incident Database, joins Chris and Dan to discuss AI safety, verification, evaluation, and auditing. They explore why benchmarks often fall short, what red-teaming at DEFCON reveals about machine learning risks, and how organizations can better assess and manage AI systems in practice. Featuring: Sean McGregor– LinkedIn Chris Benson – Website , LinkedIn , Bluesky , GitHub , X Daniel Whitenack – Website , GitHub , X Links: AI Verification & Evaluation Research Institute AI Incident Database 38th convening of IAAI BenchRisk State of Global AI Incident Reporting Upcoming Events: Register for upcoming webinars here !
Applying the AI as Normal Technology framework to legal services
written by Ross Chu and Shima Nassiri The Causal Frontier: Measurement Beyond Randomization The gold standard for determining the causal impact of a policy or product change at a company like Lyft is the A/B test (randomized experiment). By randomly assigning users to a treatment or control group, A/B tests inherently eliminate bias, providing clean estimates of the Average Treatment Effect (ATE). However, many critical business questions and large-scale initiatives simply cannot be randomized . This forces scientists to move past traditional experimentation and leverage quasi-experimental methods. We rely on non-randomized measurement in several key scenarios across Lyft: Partnerships and Policies: Assessing the incremental impact of a partnership (e.g., linking two company accounts) is often a non-randomized assignment. Since these collaborations require coordinated operational work across both companies and are typically announced or promoted broadly, this makes controlled randomization impractical. Long-Term Effect (LTE): Measuring effects that unfold over a long period, like the LTE of high prices on future rides, is typically handled by observational studies. Post-Launch Evaluation: Continuous monitoring of a policy after it has been fully rolled out requires a method that doesn’t involve costly holdout groups or degradation tests. Biased Data: In cases where pre-existing experimental data is found to have an imbalance, a quasi-experimental approach can potentially lev…
This is a brief guide to my new art project microgpt , a single file of 200 lines of pure Python with no dependencies that trains and inferences a GPT. This file contains the full algorithmic content of what is needed: dataset of documents, tokenizer, autograd engine, a GPT-2-like neural network architecture, the Adam optimizer, training loop, and inference loop. Everything else is just efficiency. I cannot simplify this any further. This script is the culmination of multiple projects (micrograd, makemore, nanogpt, etc.) and a decade-long obsession to simplify LLMs to their bare essentials, and I think it is beautiful 🥹. It even breaks perfectly across 3 columns: Where to find it: This GitHub gist has the full source code: microgpt.py It’s also available on this web page: https://karpathy.ai/microgpt.html Also available as a Google Colab notebook NEW : buy microgpt as a triptych on my art store at karpathy.art :) The following is my guide on stepping an interested reader through the code. Dataset The fuel of large language models is a stream of text data, optionally separated into a set of documents. In production-grade applications, each document would be an internet web page but for microgpt we use a simpler example of 32,000 names, one per line: # Let there be an input dataset `docs`: list[str] of documents (e.g. a dataset of names) if not os . path . exists ( 'input.txt' ): import urllib.request names_url = 'https://raw.githubusercontent.com/karpathy/makemore/refs/heads/…
Peter Steinberger is the creator of OpenClaw, an open-source AI agent framework that’s the fastest-growing project in GitHub history. Thank you for listening ❤ Check out our sponsors: https://lexfridman.com/sponsors/ep491-sc See below for timestamps, transcript, and to give feedback, submit questions, contact Lex, etc. Transcript: https://lexfridman.com/peter-steinberger-transcript CONTACT LEX: Feedback – give feedback to Lex: https://lexfridman.com/survey AMA – submit questions, videos or call-in: https://lexfridman.com/ama Hiring – join our team: https://lexfridman.com/hiring Other – other ways to get in touch: https://lexfridman.com/contact EPISODE LINKS: Peter’s X: https://x.com/steipete Peter’s GitHub: https://github.com/steipete Peter’s Website: https://steipete.com Peter’s LinkedIn: https://www.linkedin.com/in/steipete OpenClaw Website: https://openclaw.ai OpenClaw GitHub: https://github.com/openclaw/openclaw OpenClaw Discord: https://discord.gg/openclaw
This is a transcript of Lex Fridman Podcast #491 with Peter Steinberger. The timestamps in the transcript are clickable links that take you directly to that point in the main video. Please note that the transcript is human generated, and may have errors. Here are some useful links: Go back to this episode’s main page Watch the full YouTube version of the podcast Table of Contents Here are the loose “chapters” in the conversation. Click link to jump approximately to that part in the transcript: 0:00 – Episode highlight 1:30 – Introduction 5:36 – OpenClaw origin story 8:55 – Mind-blowing
The Protect What’s Human campaign will push for commonsense AI safety rules at federal and state level
From Print to Digital: Making Weekly Flyers Shoppable at Instacart Through Computer Vision and LLMs Key contributors: Prithvi Srinivasan, Shishir Kumar Prasad, Kristen Morgan, Bryan Pham, Rick Shukla, Preeti Chadha, Vipul Bahubali, Ahmad Sajedi, and Ali Maleky Introduction Grocery flyers have long been a cornerstone of retail promotions, from paper inserts in the newspaper to email blasts featuring weekly deals. As more grocery shopping shifts online, however, these static promotions haven’t kept pace with customer expectations for convenience and interactivity. At Instacart, we recognized the opportunity to transform static promotional content into interactive, shoppable experiences. In 2024, we launched grocery flyers on our platform[1] , enabling retailers to upload their weekly and monthly promotions. This enabled our customers to browse through weekly deals for their favourite retailers, providing easy ways to save. Fig 1: Sample grocery flyer Customers expect digital flyers to look and feel like the physical versions they’re used to, with the added ability to tap on items and shop directly. To deliver that experience early on, we relied on a manual digitization process. This involved drawing bounding boxes around every deal and accurately matching those deals to products to our catalog — a painstaking task that required 3–4 hours per flyer. As the feature gained traction with retailers, this manual approach quickly became unsustainable. With dozens of retailers uploadi…
How can you quantify creativity?
Recently I read Julian Mendel's article, " Mechanistic Interpretability: Peeking Inside an LLM " on Towards Data Science( https://towardsdatascience.com/mechanistic-interpretability-peeking-inside-an-llm/ ), and I became fascinated by the idea of moving beyond treating Large Language Models as "black boxes." Summary of the article's premise: The post proposes that we can understand and even manipulate an LLM’s behavior by examining its internal architecture—specifically the residual stream, attention heads, and MLP layers. It treats the model as a circuit that can be reverse-engineered to see how information is processed and stored. Key examples mentioned: World Models: How models represent internal states of games like Chess or Othello. Induction Heads: Specific attention heads that allow for in-context learning. Superposition: The phenomenon where neurons are "polysemantic," representing multiple concepts at once. Steering Vectors: Using activation addition to modify a model's behavior (e.g., making it more honest or preventing refusals) without retraining weights. Conclusion of the article: The author concludes that while manual mechanistic analysis provides deep insights into safety, reliability, and human-like cognition, the field is rapidly moving toward automated interpretability to handle the sheer scale of modern models. The Question : As someone looking to transition from a general understanding of Transformers to actually performing research or experiments in Mech…
Google adds Gemini AI-powered ‘auto browse’ to Chrome, Users flock to open source Moltbot for always-on AI, Qwen3-Max-Thinking debuts, and more!
The hidden cost of being a data-driven company is context-switching for analysts due to "quick questions." DataBot allows the data team to focus on auditing analysis instead of performing it.