Ethan

@andrewdfeldman @cerebras You should consider hiring @Electricaliee. I think it would be a mistake not to at least consider it.

.@cerebras designs and manufactures exclusively in America. 120,000 sq ft of new manufacturing capacity is coming online in the next few months. And a 10,000 sq ft clean room. We are building more capacity as fast as we can. Cerebras. The fastest AI in the world. Designed and manufactured in the USA.

@mualphaxi @Electricaliee

I have been dreaming of this day for a long time. Arena is now a book publisher, and our first volume, "Silicon" is open for preorders. It's quite unlike anything you've seen: a coffee table book capturing the ecstatic beauty of silicon technology. arenamag.com/silicon

@craigzLiszt This msg should be amplified

the odds of you building a successful company after dropping out of college are super low take fewer classes and vibe code the company on the side instead

Seems to need a prompt addition to describe how it should interpret web search outputs. “Consider the content provided below that was retrieved from an up to date web search. This should act as additional material to base your answer on, not the ONLY thing (unless it is date dependent) {web search results}”

@nicdunz Very very weird, @nicdunz can you provide a share link to the convo?

ive noticed 5.4 does things like this. 5.2 never did this.

@xenovacom @Electricaliee

NEW: Alibaba just released Qwen 3.5 Small — a family of powerful multimodal models available in a range of sizes (0.8B, 2B, 4B, and 9B parameters). Perfect for on-device applications! They can even run 100% locally in your browser on WebGPU, powered by Transformers.js! 🤯

@deepigoyal @temple @Electricaliee seems cool

We're recruiting at @temple. At Temple, we are building the ultimate wearable for elite performance athletes. A device that measures what no other wearable in the world measures, with a level of precision that doesn't exist yet. To build it, we need people who are obsessive about both the craft and the category. Engineers who are also athletes. People who will wear what they build, and hate it until it's perfect. Roles we're hiring for: 🟠 Analog Systems Engineers, Electronics Design Engineers 🟠 Embedded Systems Engineers — low-level HW bring-up, embedded signal and image processing, embedded AI 🟠 Design and Validation Engineers — sensors, actuators, battery, antenna, optics 🟠 CMF Engineers, Adhesive Materials Engineers 🟠 Sensor Algorithms Engineers — estimation theory, sensor fusion 🟠 Deep Learning Engineers — ML model development for physiological metrics 🟠 Computational Neuroscientists 🟠 BCI Engineers — real-time EEG/EMG acquisition and processing 🟠 Neural Decoding Researchers — brain activity to semantic mapping 🟠 Computer Vision Engineers — facial microexpression, subvocal muscle detection 🟠 Neuroimaging ML Engineers — multimodal sensor fusion 🟠 Last but not the least, product managers who work through Figma without needing a designer to hold their hand Important – we are building for people who push their bodies to the edge. We want to be those people, not just serve them. So only people who take fitness seriously, and have body fat <16% (men) and 26% (women) should apply. If you're not there yet but will commit to getting there in three months, you can apply too; but you'll be on probation until you are. Write to build@temple.com with your core skill as the subject line. Come find your tribe.

@nicdunz What everyone else said + prediction market participation

how does arena even make money

@reinerpope @Electricaliee

We’re building an LLM chip that delivers much higher throughput than any other chip while also achieving the lowest latency. We call it the MatX One. The MatX One chip is based on a splittable systolic array, which has the energy and area efficiency that large systolic arrays are famous for, while also getting high utilization on smaller matrices with flexible shapes. The chip combines the low latency of SRAM-first designs with the long-context support of HBM. These elements, plus a fresh take on numerics, deliver higher throughput on LLMs than any announced system, while simultaneously matching the latency of SRAM-first designs. Higher throughput and lower latency give you smarter and faster models for your subscription dollar. We’ve raised a $500M Series B to wrap up development and quickly scale manufacturing, with tapeout in under a year. The round was led by Jane Street, one of the most tech-savvy Wall Street firms, and Situational Awareness LP, whose founder @leopoldasch wrote the definitive memo on AGI. Participants include @sparkcapital, @danielgross and @natfriedman’s fund, @patrickc and @collision, @TriatomicCap, @HarpoonVentures, @karpathy, @dwarkesh_sp, and others. We’re also welcoming investors across the supply chain, including Marvell and Alchip. @MikeGunter_ and I started MatX because we felt that the best chip for LLMs should be designed from first principles with a deep understanding of what LLMs need and how they will evolve. We are willing to give up on small-model performance, low-volume workloads, and even ease of programming to deliver on such a chip. We’re now a 100-person team with people who think about everything from learning rate schedules, to Swing Modulo Scheduling, to guard/round/sticky bits, to blind-mated connections—all in the same building. If you’d like to help us architect, design, and deploy many generations of chips in large volume, consider joining us.

@lulumeservey @Electricaliee

Banger

@loganthorneloe @jonxlegasa

My team at Google is hiring ML and software engineers in Pittsburgh and Mountain View! We're looking for someone who is: - Curious - High agency - Excited about solving difficult problems If this is you, send me a DM. (If you've already sent one and I didn't respond, ping me again!)

@Angaisb_ They published smthn about extended thinking leading to convoluted thinking. Perhaps they just focus so heavily on non reasoning - instruct. Compare gpt5.2 chat or instant to sonnet no reasoning, no comparison, literally like 10-15 point difference on @ArtificialAnlys

How do Claude models get such good results while barely thinking at all? I've never seen Opus 4.6 or Sonnet 4.6 think for more than a few seconds, while every other model takes way longer

@theo Have to run a script to open them.

Skills are literally just markdown files how the hell can they have downtime???

@chatgpt21 @OfficialLoganK is this just a diff system prompt?

It’s nice to know Google is serving a much worse lobotomized version of Gemini 3.1 pro to the pro subscribers vs the 200-300$ tier in the AI studio without a formal declaration of doing so “Prompt. Make a detailed svg of a frog riding a horse. be creative.” One is in app and one is in the AI studio, can you guess which one?

@initjean This was part of the flex for Opus 4.5; it was able to use way less tokens per task.

how is Sonnet 4.6 more expensive than Opus 4.5?!!

@sean_t_strong Fire.

Our most capable Sonnet model is here. Try it out for frontier performance in coding, agents, and professional work. Honored to play a small part in shipping this alongside such an incredible team.

@alexalbert__ Nice.

Sonnet 4.6 is here. It's our most capable Sonnet model by far, approaching Opus-class capabilities in many areas. Very excited for folks to try this one out. The performance jump over Sonnet 4.5 (which was released just over four months ago) is quite insane.

Sonnet 4.6 has improved on benchmarks across the board. It approaches Opus-level intelligence at a price point that makes it practical for far more tasks.

@shadcn @Electricaliee called

Need: Add Drawing button to every prompt input. And it opens Excalidraw.

@rohanpaul_ai @Electricaliee find an opportunity

The US govt just announced $145M for apprenticeship-based training in AI, semiconductors, and nuclear energy. This as part of a push to reach 1M active apprentices nationwide. Very powerful signal that AI work is being treated like a skilled trade, not just a white-collar degree job. That includes deploying models, managing data centers, operating inference clusters, and handling the systems and hardware around them. A lot of this workforce will be trained on the job, without needing a PhD or an elite CS background. The incentive payments are “pay-for-performance,” so sponsors get paid when they create or expand apprenticeships and successfully move people through measurable milestones, rather than getting a big check up front for training activity. Apprenticeship growth often stalls because sponsors eat early costs for setup, mentoring time, and administration before they know the program will scale. The new setup funds up to 5 cooperative agreements, and the funding rules require at least 85% of dollars to flow out as incentive payments, with the incentive model proposed by applicants.

@boxmining @Electricaliee

Chinese engineers refactored openclaw in GO for hyper efficiency. Can run on your $10 dollar raspberry Pi instead of $399 Mac mini