apple shipped the touchscreen mac, the exact thing jobs called ergonomically terrible for a decade. $2,899, 24h battery. i WILL be tapping that OLED
nvidia open-sourced a transcription model that keeps every voice straight even when EVERYONE talks over each other at once. my family group chat goes in first
Meet NVIDIA Nemotron 3.5 Transcribe 🎙️
An open-weight, promptable 30B MoE model with ~4B active parameters that brings transcription in 40 languages, speaker attribution, translation, contextual prompting and audio understanding into one checkpoint.
Watch it keep track of who said what when everyone’s talking at once.
1/ 🧵
egoist wrote a whole video download engine by himself. no yt-dlp, no ffmpeg, one 6MB binary, and the cli speaks --json so agents can drive it. my agents are about to be so spoiled
Introducing Downbit: download videos from any web page, a 6MB native app, no external dependencies.
downbit.app
openai buying google's chips a week after putting its models on google cloud. pick a side man
OPENAI REPORTEDLY PLANS TO BUY GOOGLE AI CHIPS
OpenAI plans to purchase AI chips from Google ($GOOGL), according to The Information, citing a person familiar with the matter.
Financial terms and the size of the potential deal were not disclosed. Reuters has not independently verified the report.
GrapheneOS dropped experimental Linux 6.18 builds for the whole Pixel 10 line, expecting kernel panics and a dozen more crashes. got a spare pixel? go break things for them
Experimental releases of GrapheneOS with our initial port to the Linux 6.18 branch are now available for testing on the Pixel 10, Pixel 10 Pro, Pixel 10 Pro XL and Pixel 10 Pro Fold. We've done initial testing but there will likely be at least a dozen more crashes and other issues to find and fix.
People should only help with testing this if they're able to tolerate regular kernel panics and various hardware functionality not working. We've done basic testing of Wi-Fi, Bluetooth, cellular, GNSS, Vulkan and virtual machines via the Terminal app and microdroid API but there's far more to test.
If you're interested in helping with testing, please join either our #testing:grapheneos.org channel on Matrix or #testing on our Discord server. We've posted links to over-the-air updates for each of the 4 releases in both channels. These are production builds signed with the official signing keys.
Users can update to these from our current releases and update from those to future experimental and regular releases. Hardware functionality should be the focus. Bluetooth / BLE, NFC, assorted Wi-Fi APs / standards, DisplayPort alt mode, USB peripherals, USB host modes, GNSS and NFC are examples.
These experimental releases have all of the changes from our current development branch and are security preview releases with the security preview patches from our previous release. We have 4 official build machines so we can build for all 4 devices in parallel and make new releases as people test.
Do not install the experimental releases unless you can tolerate regular freezes, kernel panics and broken hardware functionality. These are highly experimental and you should not expect the device to be very usable during the early testing. We'll make new releases with fixes for discovered issues.
cmu + browser use built the honest agent test: the judge reads the files agents hand in, and bluffing scores a flat zero (one model lost 75 of 100 runs to it). best agent still only perfect on 47/100
Introducing: Odysseys2 - hardest long horizon benchmark for web agents based on real user tasks 🏗️
We collaborated with @JangLawrenceK @rsalakhu from Carnegie Mellon University
> SOTA agent fully completes only 47/100
> 5 agents complete none
> 8-12 real rubrics per task, 60 min on the live web
Web agents are becoming better than humans at knowledge work.
odysseysbench.com/v2
IT RUNS CRYSIS. this is how you launch a computer
wonder relaunched: your running app shows up on a design canvas as editable layers, edits go back to code and out as a PR. pointing this at the chat thing's ugly buttons tonight
cerebras' whole-wafer chips bring their stupid fast inference into oracle cloud. agents stop staring at a loading bar. planned 2027, so a promise for now
The fastest inference in the world is coming to @OracleCloud.
@cerebras and @Oracle partner to enable blisteringly fast inference on OCI.
Cerebras brings fast AI.
Oracle brings deep expertise building and running AI infrastructure at scale.
Enterprises have trusted Oracle with their most important data and applications for decades.
Partnering with Cerebras gives Oracle customers more choice to match each AI workload to the compute that suits it best.
Coming soon, they’ll be able to build agents with true super fast AI in the cloud they already use.
Thank you to @MThiagarajan_ and the @Oracle team.
Excited to build this with you.
Full announcement here: investors.cerebras.ai/news-r…
anthropic bought a biotech run by ONE guy. he ran 11 labs through claude and went from idea to promising mouse results in 18 months. the most solo dev thing i've ever seen
Over the past few months, Anthropic scientists have engaged in deep research partnerships with six organizations developing new medicines. We've also begun drug discovery efforts of our own.
In doing so, we’ve learned much more about how Claude can support the many stages of drug development.
In this blog post, we share some early results from these efforts, and describe how this work fits in with our wider ambitions in the life sciences.
anthropic.com/research/early…
ex-Twitter CEO Parag Agrawal's Parallel launched a decisions API: fast answers grounded in the live web, sources attached, a tenth of a cent a call, up to 100 questions in one request. i have a spreadsheet and a dream
Introducing the Parallel Decisions API: the first web-grounded decisions engine.
Decision models like Jev can answer in an instant, but they lack the recency of current news. Now you can get both in a single call.
Decisions API is live today at $0.001 per request.
Try it with 5,000 free requests: platform.parallel.ai/play/de…
Here are some examples we've been experimenting with:
- Fact-check content (@Grok is this true?)
- Refresh stale banking records
- Classify companies and leads with fresh signals
- Backtest trading strategies
- Track competitor e-commerce inventory
the devin folks had 20 real open source maintainers grade AI PRs, vague prompts and all. best model: 45%. letting the models think longer makes them score WORSE here, this eval has seen my repo
Introducing FrontierCode 2.0: our new coding benchmark that measures a model’s ability to go beyond the spec.
Most coding evals measure a model’s ability to write code that meets a spec. With FrontierCode 1.0, we asked a new question: would you actually merge the PR?
FrontierCode 2.0 is comprised of more diverse tasks, featuring larger scopes and higher mergeability standards. We are now asking: not only would you merge this PR, but should you?
One key feature of FrontierCode 2.0 is underspecified prompts. Rather than including every engineering decision, prompts are deliberately ambiguous, giving the models room to demonstrate their judgment.
The result is a benchmark that is far from saturated, with the top model scoring just 45%.
