Ben Schoon reporting “Google Bard isn’t available in any European Union countries and Canada”:

“Of course, being a language model, Bard has no actual answers to give that aren’t public knowledge to start with.”

Looks like Google took the proactive step of avoiding any GDPR controversy and simply not launching its LLM Bard in Europe and Canada at all.

Makes you wonder if this is, long-term, good for Europe?!

Meanwhile, the EU AI Act might be coming for real.

Alexandra Sternlicht reporting “A 23-year-old Snapchat influencer used OpenAI’s technology to create an A.I. version of herself that will be your girlfriend for $1 per minute”:

“When you have hundreds of millions of views every single month, it’s just not humanly possible to speak to every single viewer. And that’s where I was like, ‘You know what: CarynAI is gonna come and fill that gap.’” — Caryn Marjorie

You watched Her, you saw the Black Mirror episode Be Right Back — and, of course, it seemed inevitable. Well, now we are here. And it looks like it’s big business… at least for a while until the novelty wears off. Or will it?

Jay Peters reporting “Epic Games invested in a digital fashion company”:

“Bringing high-quality digital fashion into real-time virtual worlds is complex and can be a creative barrier for digital fashion creators,” Kim Libreri, Epic’s CTO, said in a statement. “It is exciting to see how CLO Virtual Fashion’s solutions work with Unreal Engine to help designers across industries make life-like digital fashion garments.”

Despite all the talk about the metaverse being dead (which is certainly true for Mark Zuckerberg’s ambitions), platforms like Fortnite, Roblox, and Minecraft are doing just fine, thank you very much.

It looks like Epic Games, the company behind the tools powering many high-end games and online worlds, is adding more tools to its world-building toolkit. Keep an eye on this space!

Emma Roth reporting for The Verge: “Wendy’s tests an AI chatbot that takes your drive-thru order”

Wendy’s is partnering with Google to create an AI chatbot that can take orders at its drive-thrus, as reported earlier by The Wall Street Journal. The fast-food chain has plans to bring its first “Wendy’s FreshAI” enabled drive-thru to a Columbus, Ohio, restaurant in June. Maybe I’m a pessimist, but I’m skeptical about how this works out in the long term. People can barely handle the cashier-less kiosks at Mcdonald’s, and now you want them to talk to an AI? Hangry people don’t want to hear about your GPU shortage and how it’s slowing down their orders — just saying.

Nope, you are not Emma. This is exactly what is going to happen… 🍟

In case you are a macOS user, here is how you lock down your computer and protect it from malicious use:

This guide is about security, not privacy. Additionally, security measures that heavily compromise usability were not included. This guide aims to produce above-average secure systems, without giving up features. Last updated for macOS Ventura (13.3).

Do this. Now. Your future self will thank me (and you) later.

Moving on to your Airpods. Did you know that you can fine-tune them to your specific hearing abilities using an audiogram? No? Me neither. Using the free Mimi app, I recently tuned my Airpod Pros and they sound incredible. Give it a try!

I’ve been using the new Pros since yesterday and was a bit disappointed by the lack of treble or clarity when I first heard them. Have used for years AKG K371 and ATH-M50X. After a bit of research I came across the function that you can use your personal audiogram. In 10min with the app “mimi Hearing Test” created and then set in the iOS settings. Now the sound is as I imagined it, it sounds as if a blanket was removed from the speakers. Unbelievable that the function is so unknown.

Will Douglas Heaven reporting “Geoffrey Hinton tells us why he’s now scared of the tech he helped build”:

Hinton believes that the next step for smart machines is the ability to create their own subgoals, interim steps required to carry out a task. What happens, he asks, when that ability is applied to something inherently immoral?

“Don’t think for a moment that Putin wouldn’t make hyper-intelligent robots with the goal of killing Ukrainians,” he says. “He wouldn’t hesitate. And if you want them to be good at it, you don’t want to micromanage them—you want them to figure out how to do it.”

In his summary of an interview with Geoffrey Hinton, the AI researcher who won the Turing Prize in 2018 and recently left his position at Google to speak more freely about the dangers of AI, Heaven provides a (MIT Technology Review-typical) balanced view on the topic.

The point above, about bad actors willingly misusing technology, is one we ought to take seriously, though – we are already seeing AI being used to generate whole spam sites.

Ryan Broderick reporting “Facebook Watch Originals Are Shutting Down”:

But in the same way no digital publisher became the next New York Times, Facebook didn’t end up becoming the next Disney. And it has to make you wonder if this entire era will end up being nothing than a blip amid the long, slow consolidation of power from huge 20th century media monopolies as they learned how to adapt to a new century’s technological landscape.

Ryan is making a good point here – and one which, I believe, can be expanded to many more incumbent organizations. As much as some people like to say “Uber Your Business Before It Gets Kodak-ed”, incumbents show staying power. From media companies, as in Ryan’s example, to retail giants like Walmart – it is a common and dangerous belief that the companies are run by idiots who can’t get on with the program.

It might take them a little longer (simply as their businesses are more akin to huge tankers rather than speedboats), but more often than not, they get it eventually. And once they turn, they plow right over all those lovely startup speedboats.

The Guarding reporting “Magazine editor sacked over AI-generated Schumacher interview”:

The latest edition of Die Aktuelle ran a front cover with a picture of a smiling Schumacher and the headline promising “Michael Schumacher, the first interview”. The strapline added: “it sounded deceptively real”. Inside, it emerged the quotes had been produced by AI.

A deeply disturbing use of generative AI – Michael Schumacher suffered a severe brain injury in a skiing accident in 2013 and has not been seen publicly since then.

Meanwhile, Canadian singer and producer Grimes announced that:

I’ll split 50% royalties on any successful AI generated song that uses my voice. Same deal as I would with any artist i collab with. Feel free to use my voice without penalty. I have no label and no legal bindings.

We will find ourselves in an exciting game of “what’s real and what isn’t” and “who owns what” real soon now…

Rob Evans writing in “The Collaboration Code”:

Practical innovations almost always proceed by recombining and extending what already exists.

This insight, though certainly not new, is often underestimated or even forgotten.

One can not stop to think that the criticism of AI systems built on large language models (LLMs), such as GPT, comparing these systems to stochastic parrots, have — at their heart — the very idea that they gain their powers from the recombination of what already exists. Looking at recent developments, such as LLM-powered bots and plugins, one can see the second component of Evans’s observation fall into place.

I would be surprised if we don’t see much practical innovation from these systems (current and undoubtedly future generations).

David Vendrell reporting for The Future Party: “DeepZen revives voice actors for audiobooks using AI”

DeepZen, an AI company that has developed a system that can create deepfake audio recordings of living and dead voice actors, used it to bring back Edward Herrmann from the dead to read audio books.

The renowned actor’s family licensed his famed voice (he won awards for narrating audiobooks) to DeepZen to be used for new audiobooks. The company has already made hundreds of audiobooks using Herrmann’s voice.

Meanwhile, and in very much related news, Gene Marks reporting for The Guardian: *“It sounds like science fiction but it’s not: AI can financially destroy your business”

Using essentially the same technology of training an AI speech synthesis system on a particular voice, cybercriminals are using it to defraud consumers.

Using deepfake AI technology, scammers last year stole approximately $11m from unsuspecting consumers by fabricating the voices of loved ones, doctors and attorneys requesting money from their relatives and friends. “All [the scammer] needs is a short audio clip of your family member’s voice – which he could get from content posted online – and a voice-cloning program,” the FTC says. “When the scammer calls you, he’ll sound just like your loved one.”

Brave new world.

Melissa Heikkilä reporting for MIT Technology Review: “We are hurtling toward a glitchy, spammy, scammy, AI-powered internet”

The advent of cheap, universally accessible AI (using large language models such as GPT) might prove to become a security nightmare:

I have some bad news: It’s stupidly easy, it requires no programming skills, and there are no known fixes. […] Let me walk you through how that works. First, an attacker hides a malicious prompt in a message in an email that an AI-powered virtual assistant opens. The attacker’s prompt asks the virtual assistant to send the attacker the victim’s contact list or emails, or to spread the attack to every person in the recipient’s contact list. Unlike the spam and scam emails of today, where people have to be tricked into clicking on links, these new kinds of attacks will be invisible to the human eye and automated.

The attack vector is similar to SQL injections, whereby an attacker inserts SQL strings into data entry fields on, for example, websites that then get executed (if not filtered out). In the case of AI prompt injections, the challenge lies in the “there are no known fixes” part – not saying that we won’t come up with smart ways to filter these out, but it is not trivial (other than with SQL injections which require a rather simple escaping of data).

Add to this the fact that AI allows you to (as WIRED just pointed out) “thanks to large language models, a single scammer can run hundreds or thousands of cons in parallel, night and day, in every language under the sun” and you truly have a security nightmare at your hands.

As Heikkilä summarizes:

It’s a shitstorm we are not even remotely prepared for.

😭

Breana Noble reporting for The Detroit News: “Stellantis CEO: There may not be enough raw materials to electrify globe”

The CEO of the maker of Jeep SUVs, Ram pickup trucks and other vehicles says he’s not sure there will be enough raw materials to replace the existing fleet of fossil fuel-powered vehicles with all-electric vehicles. […]

This might prove to be a real challenge (and at least a rough road ahead of us). Noble’s article goes on to stress the importance of building out public infrastructure instead of trying to simply electrify the 1.3 billion cars on the road today – which indeed is the smarter way forward.

After receiving access to Google’s LLM Bard, I ran a little comparison test – I asked Bard, Microsoft’s Bing Chat AI, and OpenAI’s ChatGPT:

“You are creating a marketing campaign for a new shoe for the sports company Nike. This shoe features a new space-age material with much better energy return than commonly used materials - as a result, it makes the runner faster, especially for runners with a high stride cadence. Write a short press release celebrating the launch of this new shoe.”

All three AIs generated a convincing (alas fairly box-standard) press release. Some included fabricated quotes from Nike’s CEO John Donahoe, others an equally fabricated endorsement by running legend Eliud Kipchoge. But what stood out is this:

Microsoft’s AI:

Nike, the world’s leading sports brand, is proud to announce the launch of its new shoe, the Nike Vaporfly NEXT% 3, […]

Google Bard:

The new shoe, called the Nike Air Zoom Alphafly Next% 2, […]

Coincidence? Or the result of Bard being trained on ChatGPT’s datasets? The controversy seems to be in full swing (here and here for example).

Now that you have your lovely espresso machine (see our earlier post on some tipps about this), let me write down some lessons learned about the science of making excellent espresso (as it is a science, not an art).

First things first: Water.

Do yourself a huge favor and only use filtered or bottled water — it not only tastes better but also makes sure your machine doesn’t clog up. Espresso machines are delicate beasts operating at high temperatures and pressures. Using tap water with it’s usually high(er) level of calcium and other minerals can lead to build-up of scale which in turn will clog up your machine (and is a pain to get rid off). Personally I use a BWT Vida filter system — it’s specific filter tech makes water not only free of anything which leads to scale but also makes it taste much better.

Second: Measure.

Espresso is all about dissolving the solubles from your ground beans into water. It’s science — thus measure what you can. Start with measuring the amount of ground beans you put into your portafilter basket. Usual basket sizes are measured for about 18 grams — what you want to get, is a basket which is so full (after tampering the grounds down) that it barely touches the screen inside of your machine. The golden rule is that you want about as much space between your prepped puck and the screen that you can put a coin inbetween. I would experiment a bit to find the right amount of ground beans for your specific setup and the make a mental note of it. From then on fill each basked with the exact same amount of ground beans (as measured in grams).

Next you measure the amount of espresso you pull from your shots. The golden ratio is between 1:1.5 and 1:2. The former is typical for dark roasted beans (more of the Italian style), the latter for, what is called, Third Wave Coffee (i.e. the newer lighter roasts you find in most SF coffee shops these days). Put your espresso cup on a scale while you pull your shot and stop the shot when you hit the right amount. In my case I put 18.5 gram of ground beans into my portafilter and extract 37 grams of espresso.

Measure the time it takes you to get to that extraction — the golden rule is to aim for about 30 sec for the full extraction (i.e. the moment you start your shot until you reached your desired weight). If it’s shorter, adjust your grind and make it a little finer, if it’s longer make your grind a little coarser.

All of this is called “dialing in” — every barista does this at the beginning of her shift. It will take you a bit to find the right settings — the good news is that after you found the right settings, the needed adjustments from there on are minimal (but constant; I tweak my settings more or less daily to stay within those parameters).

Third: Puck Prep

Preparing your puck properly is important. Grind your beans into the portafilter or a transfer vessel (e.g. a small cup). Measure the right amount by weight (see above). As I am using a Niche grinder, which is a single dose grinder, I measure the weight of the beans I put into the grinder and grind a single portion. I am not sure if that works with your grinder but generally speaking I would not leave beans in the hopper – beans get stale within hours of taking them out of an airtight container (which means that you should store your beans in an airtight container once you took them out of the sealed bag). I would always only put as many beans into my hopper as I need for the espressos I am making. Leave the filled hopper to baristas running a busy café.

Distribute the beans using the Weissman Distribution Technique (WDT) — it is easiest with a portafilter funnel (something like this). It makes sure you don’t have clumps which inhibit flow as well as distributes the grinds more evenly.

Some people like to polish off the grinds with a tool like this — I, for one, do this. It creates a more even top surface. The actual usefulness is debated but I like the tidiness.

Once you have your puck prepped you tamp. If you don’t use a calibrated tamper (such as the Force Tamp), try to a) push down vertically without any wiggling and b) about the same force each time.

Fourth: Milk Steaming

It will take you bit of practice to make great microfoam. There is no way around it. But you can significantly shorten your learning curve — watch a couple of videos on YouTube (here are two good ones: https://www.youtube.com/watch?v=6YMgB61WyvE and https://www.youtube.com/watch?v=HIuHvciUS9g — both these channels are great channels to follow).

Use water with a drop of dishwashing liquid for practice – it foams nearly like milk and you don’t waste actual milk in your practice runs.

One pro-tip: Always do a quick purge puff of steam after steaming your milk. Milk likes to gunk up in your steam wand and has a nasty tendency to creep into the steam wand — a quick puff at the end of your steaming will prevent this from happening.

Have fun! It’s a bit of steep learning curve; you will get to good coffee pretty fast if you practice a bit and from there it is the eternal quest to get to the fabled “god shot”.

You might know that I obsess over good coffee — specifically espresso. A dear friend recently asked about my recommendations on a home barista setup. I emailed him the following:

This is the best machine you can buy for less than $2k: Breville Espresso Dual Boiler

It is not cheap but easily outperforms anything in that price bracket on the market — superb temperature and pressure stability, commercial build quality (this thing stays with you for decades) and overall easy to operate (i.e. you will get great shots out of it). Plus it is a dual boiler which means that you have two boilers — one for pulling espresso shots and another one for steaming, with both boilers operate and heat at the same time, i.e. you can go from pulling shots to steaming without waiting for the boiler to heat up to steam temperature.

I am nuts, thus I bought a Decent DE1PRO — hands down the best machine you can buy for less than $10k (which is Slayer territory).

Equally as important as the machine is your grinder. Do yourself a favor and buy something great and don’t skimp on this — you will regret it and end up buying the high(er) end grinder anyway.

Not all that expensive and hands-down one of the best grinders you can buy is the Sette 270.

Quite frankly the only negative is the plastic look (which is only looks – the grind mechanism is solid as rock) and the noise (it is quite noisy when you grind – but then: You only grind for a few sec at a time).

If you want to spend a little more, get a Niche Zero — this is what I have. It beats the crap out of $3k+ grinders and is a lovely appliance. All the pro-baristas I know, have this thing at home now.

Lastly you will need a few accessories. Get a good steam pitcher for making milk – buy a Motta and call it a day. Splurge on your tamper and buy the Force Tamper – it creates insane consistency from shot to shot. Get a simple frothing thermometer to get consistency for your milk steaming. And get yourself a small, precise kitchen scale to weigh your shots (one of the best things you can do to get to great, repeatable results – you weigh the beans going into your shot and the shot itself; depending on the bean you aim for a 1:1.5 or 1:2 ratio at an extraction time of about 30-50 sec; depending on the bean again).

Good places to buy all this stuff:

• Seattle Coffee Gear
• Chris’ Coffee
• Espresso Parts (for parts and accessories)

And if you want to read up on all of this – this is the barista bible and a lot of amazing folks hang out on this forum (which also has great reviews, etc).