Learn With A Robot

3.6K posts

Learn With A Robot banner
Learn With A Robot

Learn With A Robot

@learnwitharobot

At intersection of Machine Learning/ Robotics/ Systems. Subscribe at https://t.co/6ax1VodCiC to learn STEM and robots. Tip Jar: https://t.co/oULXvrR8Jw

Belmont, CA Katılım Nisan 2009
268 Takip Edilen313 Takipçiler
Sabitlenmiş Tweet
Learn With A Robot
Learn With A Robot@learnwitharobot·
Deepseek released their frontier reasoning model called DeepSeek-R1. Benchmarks showed that DeepSeek-R1 comes very close to @OpenAI's reasoning model o1. We study how DeepSeek-R1 works in practice. We ask our Vector robot the same question we asked o1 and o1-preview. 🧵
English
2
0
1
582
Learn With A Robot retweetledi
Amy Koike 🌮
Amy Koike 🌮@AmyInMadison21·
I just finished presenting at #HRI2026, and I couldn't be more thrilled! What an incredible experience! 🎉
Amy Koike 🌮 tweet mediaAmy Koike 🌮 tweet media
English
0
1
11
283
Learn With A Robot
Learn With A Robot@learnwitharobot·
Just posted our newsletter article comparing desktop humanoids from High Torque Robotics, @boosterobotics, and @faunarobotics. Its a piece of work we have been working on for the last 3 months. Hope you enjoy reading it.
Learn With A Robot tweet media
English
0
0
2
27
Learn With A Robot
Learn With A Robot@learnwitharobot·
@AmyInMadison21 Very exciting. But once we 3D print all these blocks, what do we do with them? Look forward to reading your work.
English
1
0
1
17
Amy Koike 🌮
Amy Koike 🌮@AmyInMadison21·
I am already excited to present our work on "Elements of Robot Morphology: Supporting Designers in Robot Form Exploration" at HRI next week 🤖 Inspired by elements of graphic design or elements of architecture, we propose Elements of Robot Morphology, a framework that identifies five fundamental elements of robot form (perception, articulation, end effectors, locomotion, and structure). To operationalize (and add playfulness to!) the framework, we built MEB, a modular, 3D-printable toolkit for hands-on exploration of robot form. We also just open-sourced the toolkit: github.com/Wisc-HCI/morph… If you're attending HRI, come say hi!
Amy Koike 🌮 tweet mediaAmy Koike 🌮 tweet mediaAmy Koike 🌮 tweet media
English
1
1
6
267
Learn With A Robot
Learn With A Robot@learnwitharobot·
Great to see our work on @PetoiCamp NybbleQ come second in search results, right after the manufacturer’s page!
Learn With A Robot tweet media
English
0
0
1
52
Learn With A Robot
Learn With A Robot@learnwitharobot·
@karpathy Could you give an approximate idea of the costs involved. Like how much you burned in 30 mins. Agree that coding agents outperform... but still hard to get something done within a budget. Are we at a point where we can fire a developer and use a coding agent with the saved money?
English
0
0
1
25
Andrej Karpathy
Andrej Karpathy@karpathy·
It is hard to communicate how much programming has changed due to AI in the last 2 months: not gradually and over time in the "progress as usual" way, but specifically this last December. There are a number of asterisks but imo coding agents basically didn’t work before December and basically work since - the models have significantly higher quality, long-term coherence and tenacity and they can power through large and long tasks, well past enough that it is extremely disruptive to the default programming workflow. Just to give an example, over the weekend I was building a local video analysis dashboard for the cameras of my home so I wrote: “Here is the local IP and username/password of my DGX Spark. Log in, set up ssh keys, set up vLLM, download and bench Qwen3-VL, set up a server endpoint to inference videos, a basic web ui dashboard, test everything, set it up with systemd, record memory notes for yourself and write up a markdown report for me”. The agent went off for ~30 minutes, ran into multiple issues, researched solutions online, resolved them one by one, wrote the code, tested it, debugged it, set up the services, and came back with the report and it was just done. I didn’t touch anything. All of this could easily have been a weekend project just 3 months ago but today it’s something you kick off and forget about for 30 minutes. As a result, programming is becoming unrecognizable. You’re not typing computer code into an editor like the way things were since computers were invented, that era is over. You're spinning up AI agents, giving them tasks *in English* and managing and reviewing their work in parallel. The biggest prize is in figuring out how you can keep ascending the layers of abstraction to set up long-running orchestrator Claws with all of the right tools, memory and instructions that productively manage multiple parallel Code instances for you. The leverage achievable via top tier "agentic engineering" feels very high right now. It’s not perfect, it needs high-level direction, judgement, taste, oversight, iteration and hints and ideas. It works a lot better in some scenarios than others (e.g. especially for tasks that are well-specified and where you can verify/test functionality). The key is to build intuition to decompose the task just right to hand off the parts that work and help out around the edges. But imo, this is nowhere near "business as usual" time in software.
English
1.6K
4.8K
37.3K
5.1M
Marcos Pereira
Marcos Pereira@marcospereeira·
if you're picking a model to run on openclaw, minimax m2.5 is a solid price/quality choice currently #1 on openrouter's weekly leaderboard (#2 and #3 being kimi k2.5 and glm 5 respectively)
Marcos Pereira tweet media
English
2
0
4
1.7K
Learn With A Robot
Learn With A Robot@learnwitharobot·
@GitaGopinath Assuming you are asking from an economist point of view, Antropic is the winner in both cases. Cursor uses multiple models behind the scenes
English
0
0
0
303
Gita Gopinath
Gita Gopinath@GitaGopinath·
Claude code or Cursor?
Português
542
43
1.1K
251.4K
Learn With A Robot
Learn With A Robot@learnwitharobot·
We programmed Vector robot to give us updates on Moltbook / OpenClaw with the help of @p0. Now we will have a busy night listening to all the latest debates on this topic.
English
0
1
5
182
Learn With A Robot
Learn With A Robot@learnwitharobot·
"If I can replace ServiceNow with 3 AI agents and save $10 million per year, why wouldn't I?"
Ricardo@Ric_RTP

This is the biggest irony in tech history. Microsoft beat revenue estimates. Stock plunged 11%, wiped out $400 BILLION in market cap. Salesforce reported growth. Stock fell 5.6%. ServiceNow beat earnings. Stock crashed 11%. SAP beat projections. Stock dropped 16%. Entire software sector entered bear market territory. Down 22% from peak. These are the companies everyone said would WIN from AI. They spent billions BUYING AI companies. ServiceNow: $7.75 billion for Armis. Salesforce: $8 billion for Informatica. They launched AI products. Built AI workflows. Hired AI teams. And the market said: You're all dead. Because investors just realized something nobody wanted to admit: AI doesn't make software companies stronger. AI makes software companies OBSOLETE. Morgan Stanley: "In an environment of heightened investor skepticism, stable growth falls short of shifting the narrative." Good earnings aren't enough anymore. The market is pricing in a world where AI replaces the software these companies sell. ServiceNow CEO tried defending on the earnings call: "AI needs workflow orchestration. ServiceNow is the gateway to this shift." Market response: 11% crash. Because here's what he didn't say: If AI can write code, automate workflows, and generate apps at a fraction of the cost, why would anyone pay $50,000 per year for enterprise software licenses? The per-seat pricing model that made SaaS companies rich is getting murdered by AI efficiency. One AI agent replaces 10 seats. One prompt replaces months of custom development. One LLM call replaces entire software categories. Klarna already proved it. CEO said they pulled Salesforce out of their stack. Built everything themselves using AI. And that's just the beginning. The software apocalypse hit hardest on companies that INVESTED IN AI: Atlassian: down 12.6% Intuit: down 7.8% HubSpot: down 11.5% Zscaler: down 6.3% Meanwhile, the companies ENABLING AI made money: Nvidia: up Semiconductor stocks: surging Memory firms: rallying The divide is brutal. Hardware companies print cash. Software companies get destroyed. Because in an AI-first world, you need GPUs to build the models. But you don't need software subscriptions when the AI builds the software for you. Jim Cramer called it the "P/E multiple compression crisis." Translation: Investors don't care about earnings anymore. They care about whether your business model survives the next 5 years. And right now software business models look doomed. They're literally stuck: If they DON'T invest in AI, they fall behind. If they DO invest in AI, they cannibalize their own products. It's a death spiral with no exit. ServiceNow spent $12 BILLION on acquisitions in 2025 alone. Trying to buy their way into relevance. And yesterday the market cooked them. The craziest thing to me tho... Most software companies beat earnings. Revenue was solid. Growth was fine. But it didn't matter. Because the market stopped pricing software on what it earns TODAY. It's pricing software on what it's worth in a world where AI does the job for free. And in that world these companies are worth nothing. This is the biggest sector repricing since 2008. $500 billion in market value gone in ONE DAY. And it's not stopping. Because every company watching this is thinking the same thing: "If I can replace ServiceNow with 3 AI agents and save $10 million per year, why wouldn't I?" The answer used to be: "Because you need enterprise-grade reliability." But now? AI agents are getting reliable. Fast. Software companies just realized they're competing with open-source models that cost $0.02 per 1,000 tokens. You can't win a pricing war against free. The companies that spent BILLIONS preparing for AI are getting killed BY AI. What an irony.

English
0
0
0
64
Learn With A Robot
Learn With A Robot@learnwitharobot·
@zaidmukaddam The competition is still there... but with groq having derailed by NVIDIA, its between Cerebras and Sambanova.
English
0
0
3
481
Zaid
Zaid@zaidmukaddam·
I miss the days when Groq and Cerebras used to compete to host the latest open-source models with the highest throughput.
English
11
0
163
11.7K