This paper investigates the counterintuitive phenomenon where fine-tuning pre-trained transformer models degrades performance on the MS MARCO passage ranking task. Through comprehensive experiments involving five model variants-including full parameter fine-tuning and parameter efficient LoRA adaptations-we demonstrate that all fine-tuning approaches underperform the base sentence-transformers/all- MiniLM-L6-v2 model (MRR@10: 0.3026). Our analysis reveals that fine-tuning disrupts the optimal embedding space structure learned during the base model's extensive pre-training on 1 billion sentence pairs, including 9.1 million MS MARCO samples. UMAP visualizations show progressive embedding space flattening, while training dynamics analysis and computational efficiency metrics further support our findings. These results challenge conventional wisdom about transfer learning effectiveness on saturated benchmarks and suggest architectural innovations may be necessary for meaningful improvements.

AI chatbots can write a poem, do your homework, draft lawsuits, and maybe even take your job, if the hype is to be believed. Can they handle your investments, too? While the use of artificial intelligence in the realm of financial advice is nothing new--"robo-advisors" have been around for years, some of which use AI--the chatbot technology is rapidly becoming more accessible to individual investors. Google's Bard and Microsoft's chatbot, powered by ChatGPT and integrated into its Bing search engine, can interact with users in plain English and can engage in surprisingly human-seeming interactions. To test the investing abilities of Microsoft and Google's respective products, we challenged each one to pick two stocks--one growth stock and one value stock--and see how they did over a three-week span compared to one another as well as a human.

Every Nintendo fan remembers the game that converted them. Perhaps it was running and jumping around as Mario in an abstract, toylike playspace, thrilling at the lightness and precision of his movement. It could have been becoming hypnotised by falling Tetris blocks on the Game Boy's tiny monochrome screen, or choosing a first Pokémon, marvelling at how the little collection of fat pixels representing your chosen critter instantly assumed an imagined personality. Millions of people had their first Nintendo moment during 2020's lockdowns, moving to a virtual deserted island full of quirky neighbours in Animal Crossing. For more than 40 years, this Japanese giant of entertainment has been making video games that have shaped the tastes of the people who played them as children; there is surely no game developer working today who is untouched by its influence.

This is Work in Progress, a newsletter by Derek Thompson about work, technology, and how to solve some of America's biggest problems. Sign up here to get it every week. In 2013, researchers at Oxford published an analysis of the jobs most likely to be threatened by automation and artificial intelligence. At the top of the list were occupations such as telemarketing, hand sewing, and brokerage clerking. These and other at-risk jobs involved doing repetitive and unimaginative work, which seemed to make them easy pickings for AI.

It is the eve of the new year, and Times Square is packed with chasidim dancing amid the bright lights, and -- if you live in Hackensack or nearby -- they may be headed toward your mailbox. The dancing chasidim are the October art for Chabad of Hackensack's calendar. Like any Jewish wall calendar, it includes holidays and candle lighting times; it includes advertisements for neighborhood physicians and birthdays and yahrzeits of community members; and like the other calendars printed by ChabadHouseCalendars.com, it includes birthdays and yahrzeits for the various Lubavitcher rebbes (and the most recent rebbetzin, who also is the last one). But what makes the calendar of interest beyond the 07601 zip code is the art, which was created by Rabbi Mendy Kaminker, who heads Chabad of Hackensack, using the latest "artificially intelligent" art creation software, which synthesized the pictures based on his written prompts. "I started seeing all kinds of unbelievable pictures on Twitter" of AI-generated art, Rabbi Kaminker said.

"…AI and the fourth industrial revolution will impact every aspect of people's lives." Artificial intelligence (AI) has reached a tipping point and is already part of our daily lives. If you've used Google Maps to determine the optimal route and transport options, you would have benefitted from the AI powered predictions backed by troves of underlying data. For companies such as banks, machine learning (ML) has also been successfully used to extend loans at much lower default rates and with much greater productivity. With the onset of the COVID-19 pandemic this year, the adoption of technology in our everyday lives has only accelerated.

The conventional wisdom says we can expect a more centralized structure. The author says the conventional wisdom has it wrong.

This is the first part of "An Outsider's Tour of Reinforcement Learning." If you read hacker news, you'd think that deep reinforcement learning can be used to solve any problem. Deep RL has claimed to achieve superhuman performance on Go, beat atari games, control complex robotic systems, automatically tune deep learning systems, manage queueing in network stacks, and improve energy efficiency in data centers. I personally get suspicious when audacious claims like this are thrown about in press releases, and I get even more suspicious when other researchers call into question their reproducibility. I want to take a few posts to unpack what is legitimately interesting and promising in RL and what is probably just hype.

NHTSA released their latest draft robocar regulations just a week after the U.S. House passed a new regulatory regime and the senate started working on its own. The proposed regulations preempt state regulation of vehicle design, and allow companies to apply for high volume exemptions from the standards that exist for human-driven cars. It's clear that the new approach will be quite different from the Obama-era one, much more hands-off. There are not a lot of things to like about the Trump administration but this could be one of them. The prior regulations reached 116 pages with much detail, though they were mostly listed as "voluntary."

Moshe Yanai has been re-thinking the box for more than 40 years, inventing new configurations of the large containers in which we keep the data generated by our constantly expanding digital lives. Data is eating the world and Yanai has been staying on top of its exponential growth by applying smart algorithms (what is now popularly called "artificial intelligence") to cost-effectively store and manage data. First at Siemens-Nixdorf, then at EMC, he went against IBM, the dominant data storage player at the time, and won. Then Yanai re-configured storage twice again, with XIV which he sold to IBM in 2007, and now as founder and CEO of Infinidat, a startup valued at $1.2 billion (in 2015) which grew 144% last year. Along the way, he also figured out disruptive innovation.