mastodon.social is one of the many independent Mastodon servers you can use to participate in the fediverse.
The original server operated by the Mastodon gGmbH non-profit

Administered by:

Server stats:

297K
active users

#dl

4 posts4 participants1 post today

Weekly output: Qualcomm’s connected-home vision, BMW’s Panoramic iDrive, Zoox’s robotaxis, Delta at the Sphere, Better Offline, Red Bull Ford Powertrains

This would have been a long week with CES alone, but nine hours after my flight back from Vegas touched down at Dulles I was on my way into D.C.  for ShmooCon. How could I not? I’ve learned an enormous amount since I first covered that security conference in 2019–and this weekend’s edition was the con’s final go-round.

1/6/2025: Qualcomm’s Smart Home Vision Includes TVs That Read Your ‘Emotional State’, PCMag

I got a head start on my CES coverage with this post I filed from home after getting embargoed copies of Qualcomm’s CES announcements.

1/8/2025: BMW Unveils Smartphone-Inspired Panoramic iDrive Dashboard, PCMag

I started Tuesday by going to the German automaker’s introduction of this driver interface, which featured comedians Tim Meadows and Ken Jeong leading a schtick in which they, the BMW executives onstage and those of us in the stands had all been shrunk down to about smartphone size.

1/8/2025: Taking a Robotaxi From Amazon-Owned Zoox for a Spin on the Vegas Strip, PCMag

My second stop in Vegas, after a quick check-in at my hotel, was to this robotaxi service’s offices to get a test ride in its autonomous vehicle–where I was delighted to see that the other journalist joining me in the car would be my long-ago Yahoo colleague Dan Howley. This became my second story datelined from Vegas this week after I needed clarification on a couple of points from Zoox.

1/8/2025: Delta Tips AI-Powered Concierge, In-Flight YouTube, More at Flashy CES Event, PCMag

The Delta Air Lines keynote introduced me to the Sphere, but only after getting to that venue reintroduced me to how utterly awful Vegas traffic can be during CES–the “Tech Express” shuttle needed more than 50 minutes to crawl the 1.3 miles from the Las Vegas Convention Center to the Venetian. The production values of Delta’s event were fantastic, but the substance of the announcements didn’t live up to that standard. Two aviation journalists who have covered the airline far longer than I have, Brett Snyder and Seth Miller, came away almost entirely unimpressed.

1/9/2025: Better Offline CES 2025: Day 4 – Pt. 2, Ed Zitron

Ed, a longtime tech publicist and more recent scourge of how growth-at-all-costs tech companies are ruining the digital world, invited me to join his podcast sometime when I was at CES, and I picked the last slot available–4:30 to 5 p.m. Thursday. Skip ahead to almost the 1:11 mark to hear my conversation with Ed and fellow CES journalists Ed Ongweso, Jr. and David Roth.

1/10/2025: Red Bull and Ford are building a new F1 hybrid race car engine—first as bits, then atoms, Fast Company

I started work on this piece months ago and then kept putting it off as I had other Fast Company projects with less flexible deadlines demand my attention first, then finally wrote and filed it the day after Christmas–just in time for it to get caught up in the publication’s winter break.

Rob Pegoraroces – Rob Pegoraro
More from Rob Pegoraro
#Amazon#BMW#ces

Алгоритмы спекулятивного инференса LLM

За последние годы качество LLM моделей сильно выросло, методы квантизации стали лучше, а видеокарты мощнее. Тем не менее качество генерации все еще напрямую зависит от размера весов и, как следствие, вычислительной сложности. Кроме того, генерация текста авторегрессионна - токен за токеном по одному, потому ее сложность зависит от размера контекста и количества генерируемых токенов. Но генерация текста не всегда имеет однородную сложность, так же как мы во многом мыслим идеями, а слова произносим “на автомате”. В статье обсудим алгоритмы, позволяющие использовать эту неоднородность для ускорения.

habr.com/ru/articles/871704/

ХабрАлгоритмы спекулятивного инференса LLMВведение За последние годы качество LLM моделей сильно выросло, методы квантизации стали лучше, а видеокарты мощнее. Тем не менее качество генерации все еще напрямую зависит от размера весов и, как...

Starting the new year with a blog post on #MixtureOfExperts. Most folks I talked with had a fundamental gap and they are a game-changer in AI.
"What are #MoEs and why is it a game-changer for AI? 🤔 Find out in my latest blog post! 👉 blog.desigeek.com/post/2025/01 #AI #GenAI #DL

Amit Bahree's (useless?) insight! · An introduction to Mixture of Experts (MoE)In #AI, what are Mixture of Experts (MoEs) and how do they work?