The cybersecurity expert SwiftOnSecurity, a decade ago, wrote a parable called "A Story About Jessica" and posted it to their (now-deleted) Tumblr blog. I found it moving and insightful. The consultancy Superbloom pointed to it …

When you think of the Cannonball Run, you probably think of a group of people making the trip from New York’s Red Ball Garage to the Portofino Hotel in Redondo Beach in the shortest (and usually most dangerous) amount of time. Police encounters, lack of sleep, zero rest stops, a spare fuel tank or two, and gallons of energy drinks are usually what fuels these attempts to make the roughly 3,000-mile route in record time.

But something was lacking in the most recent successful attempt: conventional gasoline. That’s because speed wasn’t the aim of the three friends who built and drove a solar-powered car across the country in a record-setting 13 days, 15 hours, and 19 minutes.

While that’s more than 300 hours longer than the 25-hour...

Continue reading…

Early termination fees are “a bit like heroin for Adobe,” according to an Adobe executive quoted in the FTC’s newly unredacted complaint against the company for allegedly hiding fees and making it too hard to cancel Creative Cloud. “There is absolutely no way to kill off ETF or talk about it more obviously” in the order flow without “taking a big business hit,” this executive said.

That’s the big reveal in the unredacted complaint, which also contains previously unseen allegations that Adobe was internally aware of studies showing its order and cancellation flows were too complicated and customers were unhappy with surprise early termination fees.

In a short interview, Adobe’s general counsel and chief trust officer, Dana Rao, pushed...

Continue reading…

Yesterday I went on what is pretty much a nerd tour of Hanford Reactor B. Hanford is the Department of Energy facility where they made plutonium to make nuclear bombs, and Reactor B was the first reactor built during World War II to make plutonium. It is now a National Historic Landmark. The tour included a long bus ride to the site, which gives you a better understanding of just how big Hanford is. The tour then includes several hours to wander around the building, and several presentations by docents. I still can’t quite get over the fact that the public can tour a building where they used to make plutonium, which is rather radioactive and toxic.

The science and engineering behind the facility was impressive, and I also enjoyed looking at 1940s era technology and signage. I don’t if it was that era, or the facility, or what, but there were some fun signs all over the place. There were also some hidden historical signs, like all the clocks were stopped at 10:48, which was time on the night of September 26, 1944 when the reactor achieved fission for the first time. I don’t know why, but I was amused that almost every room had a “broom” station, which were pegs on the wall where a broom and dustpan were hung. I think I only saw one room where there was actually a broom in its proper station. I guess they do less sweeping now, or they have moved onto vacuum technology.

I really appreciated that you could wander around and spend your time looking at everything. Most of the areas have decent signage to indicate what you are looking at, although the engineer that I am, I would prefer perhaps a bit more detail. My one criticism of the tour was that they completed glossed over the amount of environmental contamination created at Hanford. In an intro video we watched before taking the tour, they briefly mention that waste was buried to be dealt with later. The docent at one point said they are cleaning up the site, which is an understatement. I know enough about the site to feel like they just didn’t want to admit the contamination they caused. Signs of remediation are present in the building. There were plenty of pipes that obviously had asbestos, and the pipes had clearly been abated to encapsulate the asbestos, so it would not become airborne and thus a hazard. There were dosimeters on the walls everywhere, which were clearly been analyzed frequently.

After introducing LLM-powered chat answers on Bing in February of last year, we’ve been hard at work on the ongoing revolution of search. Bing continues to be trusted by hundreds of millions of users to find information, get answers to questions, and explore their curiosity. Today, we’re excited to share an early view of our new generative search experience which is currently shipping to a small percentage of user queries.

By combining the power of generative AI and large language models (LLMs) with the search results page, Bing’s generative search creates a bespoke and dynamic response to a user’s query.

For example, if a user searches "What is a spaghetti western?" Bing shows an AI-generated experience that dives into the film subgenre, including its history and origins, top examples and more. The information is easy to read and understand, with links and sources that show where it came from or let the user dive deeper. The regular search results continue to be prominently displayed on the page like always.



This new experience combines the foundation of Bing’s search results with the power of large and small language models (LLMs and SLMs). It understands the search query, reviews millions of sources of information, dynamically matches content, and generates search results in a new AI-generated layout to fulfill the intent of the user’s query more effectively.



We've refined our methods to optimize accuracy in Bing, applying those insights as we continue to evolve our use of LLMs in search. We are continuing to look closely at how generative search impacts traffic to publishers. Early data indicates that this experience maintains the number of clicks to websites and supports a healthy web ecosystem. The generative search experience is designed with this in mind, including retaining traditional search results and increasing the number of clickable links, like the references in the results.

This is another important step in evolving the search experience on Bing and we’re eager to get feedback throughout this journey. We are slowly rolling this out and will take our time, garner feedback, test and learn, and work to create a great experience before making this more broadly available.

Please let us know what you think by utilizing the thumbs up and thumbs down icons at the top of the generative search results or click the Feedback icon at the bottom of the search results page to provide further comments.

We look forward to sharing more updates in the coming months.


 

One interesting observation is the impact of environmental factors on training performance at scale. For Llama 3 405B , we noted a diurnal 1-2% throughput variation based on time-of-day. This fluctuation is the result of higher mid-day temperatures impacting GPU dynamic voltage and frequency scaling.

During training, tens of thousands of GPUs may increase or decrease power consumption at the same time, for example, due to all GPUs waiting for checkpointing or collective communications to finish, or the startup or shutdown of the entire training job. When this happens, it can result in instant fluctuations of power consumption across the data center on the order of tens of megawatts, stretching the limits of the power grid. This is an ongoing challenge for us as we scale training for future, even larger Llama models.

— The Llama 3 Herd of Models

Tags: meta, generative-ai, llama, ai, llms