Welcome to the TechMeme right home for Wednesday, September 20, 2023. I'm Ryan McCullough today. The UK looks like it's going to make that online safety bill into law. Instacart closed up, but ARM has been trending down on the markets. Could Apple really have considered getting into the stock trading game? Is iPad finally getting a WhatsApp app? And what if you could get AI to perform better simply by giving it gentle words of encouragement? Here's what you missed today in the world of tech.
The UK's parliament has passed that online safety bill, which we've discussed in the past. It's actually been in the work since 2019, but this recent move paves the way for royal ascent and the bill actually becoming law as soon as in the coming days. Quoting tech crunch. Controversial UK legislation that brings in a new regime of content moderation rules for online platforms and services.
Establishing the comms watchdog off-com as the main internet regulator has been passed by parliament today, paving the way for royal ascent and the online safety bill becoming law in the coming days. Speaking during the bill's final stages in the House of Lords, Lord Parkinson of Whitley Bay reiterated that the government's intention for the legislation is, quote, to make the UK the safest place in the world to be online particularly for children, end quote.
Following affirmative votes as peers considered some last stage amendments, he added that attention now moves, quote, very swiftly to off-com, who stand ready to implement this and do so swiftly. The legislation empowers off-com to levy fines of up to 10 percent or up to 18 million pounds, whichever is higher, of annual turnover for violations of the regime.
The online safety, nay, harms bill has been years in the making as UK policymakers have grappled with how to respond to a range of online safety concerns.
In 2019, these efforts manifested as a white paper with a focus on rules for tackling illegal content such as terrorism and CSAM, but also an ambition to address a broad sweep of online activity that might be considered harmful such as violent content and the incitement of violence, encouraging suicide, disinformation, cyberbullying and adult material being accessed by children. The effort then morphed into a bill that was finally published in May 2021.
The proposed legislation continued swelling in scope as a grab bag of additional duties and requirements got bolted on in response to a smorgasborg of safety concerns reaching policymakers' ears, weather related to trolling, scam ads deep fake porn or most recently
animal cruelty changes within the governing Conservative Party since 2019 have also seen a succession of different senior ministers steering the legislation, including the likes of Oliver Dowden and Nadine Dorees, who in Dorees case enthusiastically pushed to speed up the application of criminal liability powers for tech CEOs.
Another major strand of controversy is focusing on the potential impact on web security and privacy as the bill hands sweeping powers to off-com to require platforms to scan message content for illegal material. A parade of end-to-end encrypted platforms and services have warned over the risks such powers pose with several well-known services threatening to exit the UK unless the bill was amended to safeguard strong encryption.
In the event the government appears to have steered out of a direct clash with mainstream messaging services like WhatsApp by fludging the encryption issue with a carefully worded ministerial statement earlier this month. But again, privacy and security experts remain watchful.
Additionally, there is concern the bill will lead to a mass age-gating of the UK Internet as web services seek to shrink their liability by forcing users to confirm they are old enough to view content that might be deemed inappropriate for miners. Wikipedias founder Jimmy Wales is among those raising concerns about the bill as an instrument of state censorship. He's attacked the government's approach as triply bad, quote, bad for human rights, bad for internet safety and bad law.
And pledged the online encyclopedia quote, will not age gate nor selectively censor articles under any circumstances and quote. Quick note that Instacart closed up 12.33% in its Nasdeck debut yesterday, valuing the grocery delivery company at just over $11 billion. Instacart as we've said was valued at $39 billion back in March of 2021, but when they first started pricing shares this month, they were only looking at an $8-9 billion valuation, so they ended up doing better than that.
Also noting that ARM, which you'll recall recently IPO, has lost nearly 10% of its valuation since its IPO, which can happen and all eyes are on clavio today, which is expected to start trading on the New York Stock Exchange probably by the time you hear these words. This is hella interesting, sources are telling CNBC that as equities soared back in the year 2020, Apple and Goldman were working on a stock trading feature for iPhones, but they shelved that project in 2022 as markets turned south.
Quote, the effort which has not been previously reported would have added to Apple's suite of financial products powered by Goldman. Apple first teamed up with the Wall Street bank to offer credit cards in 2019 and then added buy now pay later loans and a high yield savings account. The company said last month that the savings account offering had climbed past $10 billion in user deposits.
Apple was working on the investing feature at a time of zero interest rates during COVID when consumers were stuck at home and spending more of their time and their record savings in trading shares, including meme stocks like GameStop and AMC from their smartphones. Apple's conversations with Goldman began during that hype cycle in 2022 sources said their work progressed and an Apple investing feature was meant to roll out in 2022.
One hypothetical use case pitched by executives involved the ability for iPhone users with extra cash to put money into Apple shares, one person said. But as markets were roiled by higher rates and sorry inflation, the Apple team feared user backlash if people lost money in the stock market with the assistance of an Apple product, the sources said. That's when the iPhone maker and Goldman switched directions and pushed the plan to launch savings accounts which benefit from higher rates.
The status of the stock training project is unclear after Goldman CEO David Solomon bowed to internal and external pressure and decided to retrench from nearly all of the banks consumer efforts. One source said the infrastructure for building an investing feature is mostly built and ready to go should Apple eventually decide to move forward with it. Yeah, but here's the thing. Apple is, as we know, extremely concerned about people only having positive experiences with their products.
A savings account is one thing because in theory number only go up with savings accounts. But if people see like a 20% haircut in their retirement nest egg, yeah, that's a pretty negative brand association. You could run into there. It's one thing. If you lose money in your Robin Hood account in the app on your iPhone, it's another thing. If it's your Apple app and bank account and brokerage account itself, that has lost money. People have been begging for a WhatsApp iPad app for basically ever.
And maybe they're about to get their wishes granted. WhatsApp has released a beta via Apple's test flight that works natively with the iPad built using Mac catalyst after years of requests for such an app from users, quoting the verge. From what we can see in screenshots, the iPad app works exactly like you'd expect. You can connect to it by scanning a QR code the same way you'd link your account to any other device.
You'll see a list of your conversations on the left and your current chat on the right. It's pretty much the iOS app, but instead of seeing one pain at a time, you see both. It almost makes you wonder what took so long, especially when WhatsApp had WillCathcart set all the way back in January 2022 that, quote, we'd love to do it. Meta for some reason has never liked building iPad apps. There's still no iPad app for Instagram or threads.
And even the Facebook app is just a poorly zoomed version of the iPhone app. Messenger is about the only Meta app with a half decent iPad experience. But as WhatsApp becomes more central to Meta's social ecosystem and as the company tries to convince businesses to use WhatsApp for listings and customer interactions, it makes sense to bring the app to iPad as well. The WhatsApp experience has gotten a lot of attention recently.
The Mac app was updated with better group calling, HD photos and video are now supported in the app, channels are rolling out to users around the world, and Meta has done a lot of work to make it easier to use your account on multiple devices. Maybe most interesting of all Meta appears to be working on making WhatsApp a cross-platform messaging service in order to comply with EU rules, which is yet another reason to make the app available to as many users and as many places as possible.
We've been talking about how Miro lets you work closely with your team, but let me come at this from a slightly different angle. What about the client, the customer? What if you could involve all stakeholders in the design process in real time? You've been iterating on the product, you need feedback from the client, and Miro lets you do that easily. Miro's marketing journey maps helps you turn a good product experience into a great one.
On Miro's Infinite Canvas, you can lead discussions about user needs, brainstorm improvements, and improve the consistency of the experience. Conduct customer or user interviews and capture insights on sticky notes. Map the customer experience along touch points, tracking their needs, emotions, and more. Identify problems with your current product or service and then map out potential solutions.
Imagine how much you could shorten the time to launch when your team and all stakeholders are aligned in real time throughout the whole process. Empower your entire team to take ideas from better to best with Miro. Sign up today at Miro.com slash podcast, and if you use that link, your first three Miro boards are free forever when you sign up. That's MiroMIR.com slash podcast. Think of your favorite entrepreneur. They wouldn't be anywhere without trusted partners, right?
So if your commerce platform is scaring away potential customers instead of converting them into lifelong fans, you've got to try Shopify. Shopify is the e-commerce platform revolutionizing millions of businesses worldwide. Whether you're a garage entrepreneur or IPO ready, Shopify is the only tool you need to start run and grow your business without the struggle.
Shopify puts you in control of every sales channel, so whether you're selling satin sheets from Shopify's in-person POS system or offering organic olive oil on Shopify's all-in-one e-commerce platform, you're covered. All my e-commerce businesses run on Shopify, completely transformed how I run things.
Shopify powers 10% of all e-commerce in the US, and Shopify is truly a global force powering all birds, rothies, and Brooklyn, and millions of other entrepreneurs of every size across over 170 countries. Plus Shopify's award-winning help is there to support your success every step of the way. This is Possibility, powered by Shopify. Sign up for a $1 per month trial period at Shopify.com slash ride all lower case. Go to Shopify.com slash ride to take your business to the next level today.
Shopify.com slash ride. Nurellink has opened enrollment for the first human trials of its brain implant for paralysis patients. This is expected to take about six years and might only include around 10 people going to the next level. The study will use a robot to surgically place a brain computer interface or BCI implant in a region of the brain that controls the attention to move.
Adding that its initial goal is to enable people to control a computer cursor or keyboard using their thoughts alone. The company which had earlier hoped to revive approval to implant its devices in 10 patients was negotiating a lower number of patients with the US Food and Drug Administration after the agency raised safety concerns according to current and former employees. It is not known how many patients the FDA ultimately approved.
CEO Elon Musk has grand ambitions for Nurellink saying it would facilitate speedy surgical insertions of its chip devices to treat conditions like obesity, autism, depression, and schizophrenia. In May, the company said it had received clearance from the FDA for its first in-human clinical trial when it was already under federal scrutiny for its handling of animal testing.
Even if the BCI device proves to be safe for human use, it would still potentially take more than a decade for the startup to secure commercial use clearance for it according to experts. Finally today, deep-mind researchers have detailed optimization by pro-prompting, PRO and capital letters prompting. At least that's how I think you pronounce it. This is all in aid of improving LLM performance by using what they're calling meta-prompts, like telling the LLM to take a deep breath.
This has apparently helped Google's Palm2 development significantly, especially in areas surrounding math. This is going to sound wild, but what if you just gave the AI encouragement to get it to perform better? AI, they're just as fragile as we are, quoting ours' technical. Google deep-mind researchers recently developed a technique to improve mathability in AI language models like chat GPT by using other AI models to improve prompting the written instructions that tell the AI model what to do.
It found that using human-style encouragement improved math skills dramatically in line with earlier results. In a paper called Large Language Models as Optimizers listed this month on Archive, Deep Mind Scientists introduced optimization by prompting or OPRO, a method to improve the performance of large language models such as OpenAI's chat GPT and Google's Palm2.
Use new approach sidesteps the limitations of traditional math based optimizers by using natural language to guide LLMs and problem solving. Natural language is a fancy way of saying, everyday human speech. Typically, in machine learning techniques, using algorithms such as derivative-based optimizers act as a guide for improving an AI model's performance. Imagine a model's performance as a curve on a graph.
The goal is to find the lowest point on this curve because that's where the model makes the fewest mistakes. By using the slope of the curve to make adjustments, the optimizer helps the model get closer and closer to that ideal low point, making it more accurate and efficient at whatever task it's designed to do. Rather than relying on formal mathematical definitions to perform this task, OPRO uses Metapromps described in natural language to set the stage for the optimization process.
The LLM then generates candidate solutions based on the problem's description and previous solutions and tests them by assigning each equality score. In OPRO, two large language models play different roles. A scoreer LLM evaluates the objective function such as accuracy while an optimizer LLM generates new solutions based on past results and a natural language description. Different pairings of scoreer and optimizer LLMs are evaluated, including models like Palm2 and GBT variants.
OPRO can optimize prompts for the scoreer LLM by having the optimizer iteratively generate higher scoring prompts. These scores help the system identify the best solutions, which are then added back into the Metaprompt for the next round of optimization. Perhaps the most intriguing part of the DeepMind study is the impact of specific phrases on the output. Frazes like, let's think step by step, prompted each AI model to produce more accurate results when tested against math problem data sets.
This technique became widely known in May 2022 thanks to a now famous paper titled Large Language Models R0 Shot Reasonsors. Consider a simple word problem such as, BethBakes 4, 2,000 batches of cookies in a week. If these cookies are shared among 16 people equally, how many cookies does each person consume? The 2022 paper discovered that instead of just feeding a chapot a word problem like this by itself, you'd instead prefix it with, let's think step by step and then paste in the problem.
The accuracy of the AI model's results almost always improves and it works well with chat GPT. Interestingly, in this latest study, DeepMind researchers found take a deep breath and work on this problem step by step to be the most effective prompt when used with Google's Palm 2 Language Model. The phrase achieved the top accuracy score of 80.2% in tests against GSM 8K, which is a data set of grade school math word problems.
By comparison, Palm 2 without any special prompting scored only 34% accuracy on GSM 8K and the classic Let's Think Step by Step prompt scored 71.8% accuracy. Why does this work? Obviously, large language models can't take a deep breath because they don't have lungs or bodies. They don't think in reason like humans either.
What reasoning they do and reasoning is a contentious term among some, though it is readily used as a term of art in AI, is borrowed from a massive data set of language phrases scraped from books in the web. That includes things like Q&A forms which include many examples of Let's Take a Deep Breath or Think Step by Step before showing more carefully reasoned solutions.
Those phrases may help the LLM tap into better answers or produce better examples of reasoning or problem solving from the data set it absorbed into its neural network during training. Thank you for today, talk to you tomorrow.