On this tutorial, we implement a reinforcement studying agent utilizing RLax, a research-oriented library developed by Google DeepMind for constructing reinforcement studying algorithms with JAX. We mix RLax with JAX, Haiku, and Optax to assemble a Deep Q-Studying (DQN) agent that learns to resolve the CartPole surroundings. As a substitute of utilizing a completely packaged RL framework, we assemble the coaching pipeline ourselves so we will clearly perceive how the core elements of reinforcement studying work together. We outline the neural community, construct a replay buffer, compute temporal distinction errors with RLax, and prepare the agent utilizing gradient-based optimization. Additionally,…
Author: Naveed Ahmad
Nvidia’s GTC convention had every part: trillion greenback gross sales projections, graphics expertise that may yassify video games, grand declarations that each firm wants an OpenClaw technique, and even a robotic model of the beloved snowman Olaf from Disney’s “Frozen.” On the most recent episode of TechCrunch’s Fairness podcast, TechCrunch’s Kirsten Korosec, Sean O’Kane, and I recapped CEO Jensen Huang’s keynote and debated what it means for Nvidia’s future. And sure, a giant a part of our dialogue targeted on poor Olaf, whose microphone needed to be turned off when he began rambling. Even when the demo had gone flawlessly,…
The Securities and Alternate Fee has closed its investigation into electrical automobile startup Faraday Future, regardless of SEC employees on the case recommending an enforcement motion final 12 months, TechCrunch has realized. 4 sources conversant in the investigation, who have been granted anonymity to talk about the federal government case, advised TechCrunch that the SEC knowledgeable the corporate and other people concerned within the probe concerning the closure this previous week. The dismissal of the case comes amid a historic drop in enforcement actions by the SEC, which solely initiated 4 circumstances in opposition to publicly-traded corporations in its 2025…
AI coding firm Cursor launched a brand new mannequin this week referred to as Composer 2, which it promoted as providing “frontier-level coding intelligence.” Nonetheless, an X person posting underneath the identify Fynn soon claimed that Composer 2 was “simply Kimi 2.5” with extra reinforcement studying — Kimi 2.5 being an open supply mannequin lately launched by Moonshot AI, a Chinese language firm backed by Alibaba and HongShan (previously Sequoia China). As proof, Fynn pointed to code that appeared to determine Kimi because the mannequin. “[A]t least rename the mannequin ID,” they scoffed. It was a stunning revelation, since Cursor…
Welcome again to TechCrunch Mobility, your central hub for information and insights on the way forward for transportation. To get this in your inbox, join right here free of charge — simply click on TechCrunch Mobility! In case you haven’t observed, Uber is instantly all over the place, no less than in the case of autonomous automobiles. The corporate bought off Uber ATG, its in-house autonomous automobile improvement unit, again in 2020. Uber shed a lot of its moonshots — though it maintained an fairness stake in all of them — so it might deal with its core companies of…
Elon Musk lately outlined bold plans for a chip-building collaboration between his firms Tesla and SpaceX. Bloomberg reports that Musk shared his plans on Saturday night time at an occasion in downtown Austin, Texas, with a photograph suggesting that what Musk is looking the “Terafab” facility will likely be constructed close to Tesla’s Austin headquarters and “gigafactory.” Musk stated he’s pursuing this challenge as a result of semiconductor producers aren’t making chips shortly sufficient for his firms’ synthetic intelligence and robotics wants: “We both construct the Terafab or we don’t have the chips, and we want the chips, so we…
An anonymous Substack post revealed this week accuses compliance startup Delve of “falsely” convincing “a whole bunch of shoppers they have been compliant” with privateness and safety laws, doubtlessly exposing these clients to “prison legal responsibility underneath HIPAA and hefty fines underneath GDPR.” Delve is a Y Combinator-backed startup that final yr introduced elevating a $32 million Collection A at a $300 million valuation. (The spherical was led by Perception Companions.) On Friday, the startup tried to refute the accusations on its blog, calling the Substack submit “deceptive” and saying it “comprises a variety of inaccurate claims.” The Substack submit…
We simply wrapped up one other exhausting, inspiring, and chaotic Cellular World Congress in Barcelona, and I’ve been standardizing my ideas on what we noticed. In case you got here on the lookout for incremental updates to your favourite glass slab, you have been most likely disillusioned. In case you got here on the lookout for the start of the following main epoch in private know-how, MWC 2026 was a goldmine. For years, we’ve been predicting the collision of AI, robotics, and personalization. We’ve been ready for the second when our “sensible” gadgets cease performing like extremely reactive dictionaries and…
Shortly after Amazon CEO Andy Jassy introduced AWS’s groundbreaking $50 billion funding cope with OpenAI, Amazon invited me on a personal tour of the chip improvement lab on the coronary heart of the deal, at (principally*) its personal expense. Industry experts are watching Amazon’s Trainium chip, created at that facility, for its implications for lower-cost AI inference and, probably, a dent in Nvidia’s close to monopoly. Curious, I agreed to go. My tour guides for the day have been the lab’s director, Kristopher King (pictured under proper) and director of engineering Mark Carroll (under left), in addition to the staff’s…
This week, a subject that has been boomeranging round Silicon Valley bounced into the highlight: AI tokens as compensation. The concept is simple sufficient — quite than giving engineers solely wage, fairness, and bonuses, corporations would additionally hand them a funds of AI tokens, the computational items that energy instruments like Claude, ChatGPT, and Gemini. Spend them to run brokers, automate duties, crank by code. The pitch is that entry to extra compute makes engineers extra productive, and that extra productive engineers are value extra. It’s an funding within the individual holding them, is the concept. Jensen Huang, the leather-jacket-wearing…