Tech Brew Ride Home - Wed. 03/22 – Google Bard Joins The Fray

Episode Date: March 22, 2023

Well, we’ve got Google Bard. We’ve got a new Copilot from GitHub. Bing Image Creator is rolling out. Mozilla AI launches. The low hanging fruit that is plugging AI into NPCs to make games more realistic. And at the very end, some actual non AI news! Links: Google opens early access to its ChatGPT rival Bard — here are our first impressions (The Verge) Microsoft’s GitHub to Add OpenAI Chat Functions to Coding Tool (Bloomberg) Microsoft brings OpenAI’s DALL-E image creator to the new Bing (TechCrunch) Mozilla launches a new startup focused on ‘trustworthy’ AI (TechCrunch) Ubisoft’s new AI tool automatically generates dialogue for non-playable game characters (TechCrunch) GPT-4 and professional benchmarks: the wrong answer to the wrong question (AI Snake Oil) Amazon is flooding the zone with new TVs as it crosses 200 million Fire TV devices sold (The Verge) Learn more about your ad choices. Visit megaphone.fm/adchoices

Welcome to the Tech Meme Right Home for Wednesday, March 22nd, 2020. I'm Brian McCullough today. Well, we've got Google Bard rolling out. We've got a new co-pilot from GitHub. Bing Image Creator is rolling out. Mozilla AI launches, the low-hanging fruit that is plugging AI into NPCs to make games more realistic. And at the very end, some actual non-AI news. What is this? The 20th century. Here's what you miss today in the world of tech. Look, y'all, at our neck of the woods, there's just not much other news. news at the moment. So once more into the AI breach, my friends. Yesterday, Google released Bard, its chat GPT competitor, available via waitlist in the US and UK. The general consensus I was able to glean from early users overnight was that it was quick and fluid to use,

but seemed maybe not quite as good as what you can already do with Bing, maybe more constrained. This was the hands-on take from James Vincent at the verge, quote, Like OpenAI's chat, GPT, and Microsoft's Bing chat bot, BART offers users a blank text box and an invitation to ask questions about any topic they like. However, given the well-documented tendency of these bots to invent information, Google is stressing that BART is not a replacement for its search engine, but rather a, quote, compliment to search, a bot that users can bounce ideas off of, generate writing drafts, or just chat about life with.

In a demo for The Verge, Bard was able to quickly and fluidly answer a number of general queries, offering adenine advice on how to encourage a child to take up bowling. They answered, take them to a bowling alley, and recommending a list of popular heist movies, including the Italian job, the score, and heist. BARD generates three responses to each user query, though the variation in their content is minimal, and underneath each reply is a prominent Google it button that redirects users to a related Google search. As expected, trying to extract factual information from Bard is hit and miss. Although the chatbot is connected to Google search results, it couldn't fully answer a query on who gave the day's White House press briefing. It correctly identified the press secretary

as Corinne Jean-Pierre, but didn't note that the cast of Ted Lassau was also present. It was also unable to correctly answer a tricky question about the maximum load capacity of a specific washing machine, instead inventing three different but incorrect answers. Repeating the query did retrieve the correct information, but users would be unable to know which was which, without checking an authoritative source like the machine's manual. And how does BART compare to its main rivals chat GPT and Bing. Certainly faster than either, though this may be simply because it currently has fewer users,

and seems to have as potentially broad capabilities as these other systems. In our brief tests, it was also able to generate lines of code, for example, but it also lacks Bing's clearly labeled footnotes, which Google says only appear when it directly quotes a source like a news article, and seemed generally more constrained in its answers, end quote. Then this morning, Microsoft's GitHub announced co-pilot X, adding chat and voice features to help coders with certain tasks, powered by OpenAI's GPT4, available via a waitlist as well.

Quoting Bloomberg, the new version announced Wednesday is called Copilot X, which GitHub chief executive officer Thomas Domke said he demonstrated to one of his children by asking it how to program a snake game in Python. The chat window can provide explanations of what segments of code are meant to do, create ways to test the code and propose fixes for bugs. developers can also give instructions or ask questions using their voice. The company plans to use different AI models for different tasks. The code completion features, which demand an AI that's speedy in order not to interrupt a developer's programming flow,

will keep using older technology that optimizes pace rather than perfect accuracy. The chat features will use the newer GPT4, which OpenAI says has higher accuracy rates, Stomke said. Developers can sign up for a wait list to preview the new service. Domki said he is hopeful the software can be used. used for education, end quote. And Microsoft also began rolling out Bing image creator to the Edge web browser yesterday, powered by OpenAI's Dolly and accessible via Bing Chat, letting users create images from text prompts, quoting TechCrunch. The right prompts will generate the now familiar square of four high-res dolly images. There's one major difference, though. There will be a small

Bing logo in the bottom left corner. The early Bing AI release was missing a few guardrails, but Microsoft quickly fix those. The company is clearly hoping to avoid these issues with this release. In addition to the new image generator, Bing is also getting two additional new search features, visual stories, and knowledge cards 2.0. Bing's previous version of these knowledge cards was akin to Google's knowledge panels, but they now come with the added twist of including AI-generated infographics and more interactive elements like charts, graphs, and timelines. Visual stories, meanwhile, are more akin to AI-generated Instagram stories. Best I can tell, Microsoft has been testing these quite a bit in recent weeks since they've been popping up in my searches with some frequency.

These stories come with AI-generated audio narration and depending on the topic, videos. Sometimes they will be the top search results, but they can also appear in the knowledge cards. When searching for Portland, Oregon, for example, Bing will highlight stories like how Portlander uses fungus in the city and Portland, America's quirky city. Sadly, there doesn't seem to be a way to directly link to these stories, end quote. Mozilla has announced Mozilla AI, because why not? Let's all get on board here. An AI-focused startup with $30 million in seed funding from the Mozilla Foundation to work on open source and what they're calling trustworthy AI, quoting TechCrunch. Called Mozilla AI, the newly forged company's mission isn't to just build any AI. Its mission is to build AI that's open source and trustworthy, according to Mark Sermon, the executive president of Mozilla and the head of Mozilla AI, funded by a $30 million. seed investment from the Mozilla Foundation, Mozilla's parent organization. Mozilla AI is a wholly owned subsidiary of the Mozilla Foundation, much like the Mozilla Corporation, the organization responsible for developing Firefox and Mozilla Ventures,

the Mozilla Foundation's VC Fund. Its managing director is Moez Dreyev, who previously was the chief scientist at Huawei's NOAA's ARC AI Lab and the global chief scientist at consulting company Cap Gemini. Harvard's Karim Lakhani, Credo's Navrina Singh, and Sermin, will serve as Mozilla AI's initial board members. Akani is the chair and co-founder of the Digital Data and Design Institute at Harvard, while Singh is a member of the U.S. Department of Commerce's National AI Advisory Committee, which advises the president on a range of ethical AI issues.

Sermon describes Mozilla AI as part research firm, part community, a startup dedicated to helping create a trustworthy, independent open-source AI stack. Initially, Mozilla AI's priority will be building a team of around 25 engineers, scientists, and product managers, to work on what, again, they're calling trustworthy recommendation systems and large language models along the lines of OpenAI's GPT-4.

But the company's broader ambition is to establish a network of allied companies and research groups, including Mozilla Ventures-backed startups and academic institutions that share its vision, and quote. Not even gaming is immune to this Cambrian AI explosion. One of the biggest promises for gaming, at least the seeming low-hanging fruit in terms of injecting AI into games,

is making non-playable characters more intelligent, more realistic. Ubisoft, to that end, unveiled Ghost Rider, and says this internal AI tool will help scriptwriters save time and create more realistic non-player character interactions, quoting TechCrunch. The company says the tool isn't replacing video game writers, but instead will alleviate the task of writing barks, which are the phrases or sounds made by NPCs during a triggered event.

Ghostwriter generates first drafts of barks in order to give scriptwriters more time to focus on the general narrative, crowd chatter and barks are central features of player immersion in games, MPCs speaking to each other, enemy dialogue during combat, or in exchange triggered when entering an area, all provide more realistic world experience and make the player feel like the game around them exist outside of their actions. Ubisoft wrote in a blog post.

However, both require time and creative effort from scriptwriters that could be spent on other core plot items. Ghostwriter frees up that time, but still allows the scriptwriters a degree of creative control, end quote. The process starts with scriptwriters first creating a character and a type of interaction or utterance they would like to generate. Ghostwriter then generates variations that the scriptwriter can choose from and edit to fit their needs. The process uses pairwise comparison as a method of evaluation and improvement, which means that for each variation generated, the tool provides two choices, which will be compared and chosen by the scriptwriter. Once one is selected, the tool learns from the preferred choice.

The idea is that after thousands of selections are made by scriptwriters, the tool will become more effective, and accurate, end quote. Okay, so maybe I'm wrong. Maybe that's the lowest of the lowest hanging fruit. Just a tool. Not what I'm dreaming of, which is, surely, right, we're days away from someone plugging one of these LLMs into, I don't know, the back end of Grand Theft Auto somehow. Like, interacting with these bots in a web browser already feels like interacting with the smartest NPC ever made, right? So do that thing where you instruct the bot to pretend to be a specific persona, plug it into a game, throw some rails on it so that the bot thinks it's only in the world of the game. And we're there, right?

Some version of Westworld. One more real quick, just to give you multiple sides to all this, as I continue to try to do, a substack called AI Snake Oil says that OpenAI might have violated a cardinal rule of machine learning by testing GPT4 on its training data. I believe that would be, you know, training it on itself. to evolve the next generation. Quote, To benchmark GPT4's coding ability,

OpenAI evaluated it on problems from code forces, a website that hosts coding competitions. Surprisingly, Horace He pointed out that GPT4 solved 10 out of 10 pre-2020 problems and 0 out of 10 recent problems in the easy category. The training data cutoff for GPT4 is September 2021. This strongly suggests that the model is able to memorize solutions from its training set,

or at least partly memorize them, enough that it can fill in what it can't recall. As further evidence for this hypothesis, we tested it on CodeForces problems from different times. In 2021, we found that it could regularly solve problems in the easy category before September 5th, but none of the problems after September 12th. In fact, we can definitively show that it has memorized problems in its training set. When prompted with the title of a Code Forces problem, GPT4 includes a link to the exact contest where the problem appears, and the round number is almost correct. It is off by one. Note that GPT4 cannot access the internet, so memorization is the only explanation. Because of OpenAI's lack of transparency,

we can't answer the contamination question with certainty, but what's certain is that open AI's method to detect contamination is superficial and sloppy, end quote. Again, this is beyond my ken, so read the whole thing for more gripes about GPT4 and alleged sloppiness, if that's your bag. Finally today, wow, look at this, an old-fashioned hardware news item. Amazon has announced new fire TVs, has launched a cheaper TV lineup, says over 200 million fire TV devices have been sold thus far, and rolled out Luna in the UK, Germany, and Canada. Quoting the verge, Amazon announced today that it has sold a total of over 200 million fire TV devices. That number is a combination of fire TV streamers, third-party TVs that run fire TV software,

and the company's own televisions that debuted in 2021, which, with new models introduced last year. The last update came at CES 2022 when Amazon said it had crossed 150 million sales. Alongside that news, Amazon is introducing new TVs at the top and bottom of its lineup. It's adding new sizes of the flagship FireTV Omni Q-led series and launching a new cheaper TV lineup called the Two Series. These smaller-sized budget sets are limited to HD resolution, but the entire pitch here is that you're getting the Fire TV experience built in for prices that start at just $199. The two series models technically support HDR-10, HLG, and Dolby Digital Audio, though I wouldn't expect that their panel brightness can really showcase HDR.

The Fire TV Omni Q-led series should do a better job at that, since it includes full-array local dimming. Amazon previously offered its best self-branded TVs in 65-inch sizes. Now it's adding smaller 43, 50, and 55-inch options. The Fire TV Omni Q-Clead sets support Dolby Vision, and include a room sensor that also enables support for Dolby Vision IQ and HDR10 Plus adaptive. Both of these modes automatically adjust brightness based on your environment. One of the new tricks that Amazon added to the Q-led series was the Fire TV ambient experience. You can have the TVs display artwork, various widgets, and information like your calendar, reminders, and sticky notes.

Amazon says 98% of Fire TV Omni Q-led owners are using the ambient features and that they've boosted usage of the TVs overall, so it's expanding those today with hundreds of new artificial. art work options from institutions like the National Galleries of Scotland, National Museum Wales, and Munich's Pinocotech Der Modern. Later this year, Amazon will roll out dynamic artwork that can adjust to reflect factors like nearby weather, outdoor temperature, time of day, and more. Amazon collaborated with artist Samuel Stubblefield on this feature. The new models of the Fire TV Omni Q-LED series are up for pre-order at Amazon today, starting at $449.99, and they'll be

available from best buy.com on May 11th. The entry tier Fire TV 2 series comes in 32 inch and 42 inch sizes and is available starting at $199.99 now. Rounding out today's announcements, Amazon says the company's Luna Cloud Gaming Service is coming to the UK, Germany, and Canada. Meanwhile, the FireTV Omni Q-led 4 series and 2 series are headed to the UK, Germany, and Mexico, end quote. Last night was Kid's Night on Broadway, so we took the kids to see Hamilton for their first time. They've watched it a ton on Disney Plus, but this was their first time seeing it in person. Funny thing, I guess Penny had never paid close enough attention to the second half of the show, because when she realized last night that Hamilton was cheating on Eliza, when he kissed that lady in the red dress,

she let out a curse word, the S word, followed by Hamilton, no, which caused a ripple of laughter through the crowd. And it's funny, too, I've seen Hamilton live three times now and at least half a dozen times on TV, and somehow it makes me cry at the end every single time. reliably, every freaking time. Nothing else makes me cry so consistently. And I don't even know why. I don't even find the plot of the show that profound or emotional, but that last third of the play, I'm just a mess. Anyway, good show. They did some special things for the kids last night, a sing-along, lots of fun. Talk to you tomorrow.