LLMs like OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Bard are undeniably impressive. But as the name implies, these large language models require massive computing power. The more data used to train them, the larger these models balloon.
However, most use cases don’t actually need all that sophistication. The hefty size means LLMs live in the cloud, needing constant internet connectivity. This introduces privacy concerns regarding how user data is leveraged to refine models or even stored.
In contrast, smaller models like Microsoft’s 2.7 billion parameter Phi-2 provide extensive capabilities while fitting on embedded devices or phones. Microsoft claims Phi-2 matches or exceeds models 5x its size on certain benchmarks. Running locally enables lightning fast responses, which will only improve as chipmakers optimize for AI. Local processing also allows you to keep your data private, even personalizing an SLM with your own data.
Both LLMs and more compact SLMs like Phi-2 play key roles. But model selection matters hugely, impacting project feasibility and costs. A blended approach with specialized models handling discrete tasks may offer the best solution for many companies.
P.S. I suspect this hybrid route may be what Apple wants to pursue – storing a small LM directly on devices while still enabling cloud connectivity when needed. This would allow Apple to incorporate AI while upholding their privacy commitment, even if it means less training data than the largest LLMs leverage. Their new open source tool for running models locally hints at this focus. With a multi-model approach prioritizing on-device AI, they could query the cloud only for complex requests beyond a local model’s scope.
Some of the power of EVs lies in those massive batteries – something Ford has already pitched as a whole house backup generator in emergencies. Ford’s FordPass app already allows F-150 Lightning owners to manage how much of their battery is made available during outages and notifies them when a switchover occurs.
But why stop there? Most of us commute with a predictable daily battery usage – we could allow the remaining charge to work part-time as a home power source. The batteries could charge overnight when electricity rates are lowest, then discharge during periods of peak demand reducing demand on the grid during snow storms or heat waves, all while offering homeowners a lower monthly bill.
As we shift to EVs we’re moving to a world where many US households have giant batteries that could enable many untapped opportunities for them to become crucial resilience assets when we need them most. As we build out a modern renewable energy grid mixing batteries, solar power, and other sources, our cars can play a key role – an opportunity that utilities could leverage by helping EV homeowners install two-way chargers.
In a year where states like Texas face winter and summer rolling blackouts, which are becoming more common with climate change, we should explore ways our homes can better support demand response when the grid is maxed out. For example, the average US home uses about 33 kWh of electricity per day—so a Ford F-150 Lightning with a standard 98 kWh usable capacity (the long-range trucks have 131 kWh in usable capacity) could hypothetically power a home like normal for over 3 days. A Tesla Model 3 with a usable capacity of 57.5 kWh would last around 1.5 days. Add in solar panels, additional batteries, or a household with two EVs and it’s easy to imagine homes resilient enough to weather prolonged outages, and I would much rather be the house at the end with AC and the lights on.
I upgraded to a 5G-capable iPhone model a few years ago, but decided to keep using 4G LTE rather than pay extra for 5G service. A few weeks later, I discovered Verizon had removed tethering support on my plan, essentially forcing an upgrade if I wanted to keep using that feature. From that point on, my in-home and office service and battery life were noticeably worse with 5G enabled. Calls would drop, data was slow, and I’d be lucky to get half a day of regular use before needing to recharge—likely due to the poor signal penetration into my home.
Eventually, I turned on WiFi calling as a workaround to maintain basic phone functionality without relying solely on the flakey mobile data connection. Over time, Apple has improved power optimization on newer models, so it is no longer as big a drain on battery life, but service in buildings still feels worse.
The 5G experience has disappointed me in other ways too. I travel frequently, and hotel WiFi is often so poor that I’ll use my phone’s data instead as a backup. Like clockwork, by day two, my service gets throttled to the point my device is unusable until the next billing cycle, but isn’t this supposed to be the promise of 5G? Yet these same companies suggest I ditch fiber or cable for in-home 5G? No thanks.
To me, 5G feels like a joke and a money grab since launch. If network providers truly want mainstream adoption, then become a “dumb pipe”—let customers connect any device they want under a single service plan, like plugging a phone line into your home. The early Internet boom happened because people could experiment freely on phone networks. Today, artificial limits on approved hardware and uses only restrain innovation.
So, let’s cut to the chase. Google’s AI releases have been, well, a bit disappointing. Many thought they had cooked up something close to sentient AI, given all the data they’ve gulped up for Google Search. It set them up to be the ones to watch for a groundbreaking AI moment. But honestly, Bart hasn’t quite been living up to the hype—at least when sized up to the competition.
We’ve all been on our tiptoes for ‘Gemini,’ which has been hailed as the big breakthrough we’ve all been waiting for. Yet whispers of delay have been circulating. Then, out of nowhere, Google rolls out a preview of what has been in the works, and it’s a reminder that they’re in the AI game for the long haul, not just for quick wins.
Watch the video. For me, the real gem is what appears to be a multi-model design where several models play in tandem, stitching together responses with a common thread of understanding. It mirrors how we humans integrate our senses—sight, scent, hearing, touch, and taste—to form a singular, cohesive output.
I’m looking forward to seeing how ‘Gemeni boosts what we get out of Bart and how it shows up in the model garden for GCP (Google Cloud Platform).
I thought personalized chatbots would be the biggest drop from Meta, but they also released an image generator named Imagine that rivals OpenAI’s DALL-E and other image generation AI tools like Midjourney and Stable Diffusion. I may be in NYC vs. Mars, but it does a pretty darn good job, check it out.
Meta’s showing everyone it’s got some serious AI muscle with genuinely interactive chatbots, all building on Meta’s AI models like Llama and EMU for video editing. Now, the much-anticipated Meta AI characters have arrived, complete with unique personalities, ready for a chit-chat about anything that’s on your mind.
But hold onto your hats, because here’s the wild part:
“In addition, the company said more of its AI characters will support search powered by Bing and it will begin experimenting with “long-term memory” in several — meaning, the characters will learn and remember your conversation when it’s over.”
We’re stepping out of the ephemeral chat bubble and into a world where recollections create connections. Long-term memory in chatbots? It’s a brave new world of continuity and context, folks.
The fine print in those terms we breezily accept turns out to be a bigger deal than we thought—it’s giving us a lease on our digital stuff, not real ownership. When you pick up a digital movie, music, or game, it’s all tangled up in some complex licensing deal that could make your stuff just stop working or disappear.
Remember when owning a CD or a DVD felt solid? Well, even those good old physical copies are no longer a sure bet. Today’s video games are a prime example; they land in our hands unfinished requiring huge downloads and bug us with updates and digital checks just to play.
And it’s not just entertainment. Our houses are filled with ‘smart’ things—lights, locks, and other gadgets—that rely on costly online services. And we’ve all seen what happens when those services go dark—the stuff we paid for turns into expensive bricks.
So what’s really yours today? If the internet suddenly went poof, a lot of our tech would just be expensive paperweights. I guess you could always curl up with a good book—just make sure it’s not on an e-reader.
It’s obvious at this point that the future of gaming is not consoles but subscription-based gaming platforms, each with tentpole or exclusive content. We have long proved that even the heaviest games can stream over the Internet and run on limited terminals without the bulk and power of expensive consoles with features like Xbox Remote Play.
These days, Netflix features games above most shows and movies. Xbox, PlayStation, and Nintendo offer monthly “game passes” full of content and quarterly exclusive drops. Apple long ago moved to subscription games with its arcade product.
The big question is how long before the Apple AppStore and Google Play store concede to allow these game platforms a place – even if it directly competes with them or erodes the stickiness within their ivy walls – to provide a native gaming experience for the big console companies. After all Microsoft has tried and failed to offer a first-party native app.
As flawed as Epic’s lawsuit has felt, this seems like the real issue and heart of the coming war unless gaming companies and platforms can find a way to work together.
Popcorn lovers rejoice the shortage is over as the long OpenAI drama that began with the sudden firing of CEO Sam Altman, two interim CEOs, Microsoft starting a new division and hiring Altman, has finally ended with a new board and Altman back as CEO. If you plan to skip the link article, this is the most significant bit:
“In a letter circulated internally at OpenAI and subsequently published to the OpenAI blog, Altman announced that Mira Murati, who was briefly appointed interim CEO by the previous board, will return to her role as CTO, and confirmed that the initial new board will consist of Bret Taylor, the former co-CEO of Salesforce; Quora CEO D’Angelo, who served on the previous board; and economist and political veteran Larry Summers. Microsoft will also gain representation on the board in the form of a non-voting observer.”
I’m a bit of an oddball, constantly running beta versions of iOS (and Android) year-round. So, when the latest release hits, the new features are old news to me. However, every now and then, something I’ve been using nonchalantly catches me off guard—in this case, Apple’s new NameDrop feature, which I happen to love.
NameDrop lets you effortlessly share your contact information with another iPhone by bringing two unlocked phones in close proximity. It’s a far superior alternative to the awkward dance of scanning each other’s LinkedIn QR codes at a networking event. You have control over how much information you share, allowing you to conceal personal emails or phone numbers. Plus, you can even choose a poster screen or a picture to represent yourself.
Interestingly, some police departments got a bit jittery about the idea of strangers getting hold of your personal details, but truth be told, it’s quite safe and not much different from exchanging a business card. If the paranoia still lingers, the linked article provides details on how to turn it off.
Give it a shot, especially if you often find yourself out and about, navigating the networking scene.
For my 2023 tech resolution, I was determined to free my pockets from the jingle-jangle of keys, and folks, I’m thrilled to announce that I’m now proudly keyless. I kicked things off by upgrading the locks on my house, opting for smart locks like Yale and Level+—they still play nice with a physical key but graciously respond to Apple’s Home Key, activated with a simple tap of my watch or phone.
The car key situation seemed like a tougher nut to crack, even with more cars embracing mobile app unlocking and Apple teasing limited car key support. But lo and behold, Tesla’s keyless control, automatically unlocking when I’m in close proximity, turned out to be the game-changer I needed.
Look ya’ll, it’s time to bid farewell to the era of keys. Digital keys aren’t just the future; they feel like the future, and when they do their thing, it’s downright magical. No more fumbling through pockets, bags, or purses—stroll up, and voila, it knows and unlocks. I’m eagerly anticipating the day when standardization lets all of us drop those stabby metal objects from our pockets. Bring on the digital key revolution!
Sam Altman is back as CEO of OpenAI after four board members of the nonprofit that runs the AI company attempted a poorly planned coup. Much of this drama played out over an intense Thanksgiving holiday — and if you missed it, I highly suggest you check out last week’s newsletter.
As the OpenAI coup story ends (or begins), one huge question remains: What did the OpenAI board learn that spooked them enough to destroy one of the most important companies and innovations ever created? Some think it may have been an internal project named Q* (pronounced q star), led by now-former board member and Chief Science Officer Ilya Sutskever, that allows AI to solve math problems.
If so, here’s my take: AI as we know it is amazing but falls short of being considered a true AGI or an artificial general intelligence. Many current models are LLMs or large language models that can predict the next word in a string, but they don’t truly think or solve logic problems on their own – they’re reliant on answers that already exist. Solving math and solving it in novel ways, takes great problem-solving ability that this q* project may have shown.
If that’s the case, then a well-funded group could build an AI that can access and interpret more data than a human ever could with the processing power to solve unknown math problems. A system that can do that could solve cryptographic and encryption problems that require quantum computing.
This is all conjecture and rumor for now, but something clearly spooked this board enough to take potentially career-ending measures to slow down the progress of AI.
Maybe we’ll eventually hear the full OpenAI story and what truly sparked such dissent. But one thing is clear – AI has the potential to be the most important innovation since electricity, a stunningly transformational breakthrough that could change the course of mankind. The question is, should we tap the brakes or push forward full speed ahead? I say keep the pedal to the metal, what say you?
Executives get the axe all the time, but these are usually very coordinated events with press releases, company all hands, and well-crafted letters to customers and partners. Open AI did none of that, leaving us all assuming the worst.
So far, we know Sam Altman, CEO and Co-founder, was fired on the same day Greg Brockman, the other co-founder and board member, resigned from the board. That could be because of a disagreement, but the OpenAI board felt it necessary to announce this 30 minutes before markets closed while giving its billion-dollar partner Microsoft only a 5-10-minute heads-up.
Maybe in a week or two, between leaks and insider journalist reports, we’ll have some idea, but this is a huge decision for a high-rolling start-up (even if it’s a non-profit) that has the entire world watching. I hope the “crime” fits the punishment.
In the meantime, for those of us using OpenAI’s products, does this shake your confidence in the company? I’m personally cautious, but I want to know if this is due to more significant security, copyright, or privacy issues.
The nicer the hotel shower, the less functional it appears to be. I travel and have traveled a lot in my career, and the hotel bathroom is consistently bad, with a shower or tub that feels designed to create a flood.
Until reading this article, I thought I was alone with my thoughts on the many confusing showers I’ve used, but it looks like I’m not. The worst are half-glass or partially enclosed showers that for some reason all suffer from a weirdly placed drain or a basin that is tilted to flood the bathroom floor.
While we’re talking hotel bathrooms, I hate the new trend of giant bottles of soap, shampoo, and conditioner glued or bolted to the wall so they’re cleaned by the shower’s steam. Those things suck.
Meta released a blog on its internal research products Emu Video and Emu Edit, two new forms of generative AI that can create new videos or edit videos and images from a prompt, and it’s impressive. The link below limits us to preselected prompts but demonstrates how much is possible with just a thought or idea. Check it out!
Four years ago, I switched to Duck Duck Go as my default search engine, and it does a fantastic job for 95% of my searches. For the last 5% of searches, I try Google, and I’m often surprised to find its results to be about the same and, in several cases, the overall Google experience to be much worse.
Maybe I feel jaded by Google’s shift from clear, concise results to a revenue-focused devolution of its UX, with the upper third of search results covered in ads, tons of sponsored links, giant login boxes, and unhelpful shopping suggestions, but I find it hard to say Google is the truly superior product.
Your view may differ, but the heart of Mike Masnik’s article is how you would punish Google if it is found to be a search monopoly, especially if its fault is building a product that’s so good others can’t match it.