• No One Knows How to Test AI Anymore
  • Everyone Is in Love with the Samsung Galaxy Z Fold8
  • Anyone Know What's Going on At Google?

No One Knows How to Test AI Anymore
Honestly, the title is probably misleading. I think the proper statement is that no one ever knew how to test AI properly. But for years this wasn't a pressing issue because the AI being tested wasn't capable enough to expose the underlying human incompetence or exploit the holes in the walls being used to contain it. Now AI models and agents are running amok on the internet, even creating their own message boards to help support their efforts. Previously in this column, I noted that AI companies need to take testing their AI models more seriously. You shouldn't need more than one example of an AI model escaping its supposed sandboxed environment to understand that a better sandbox is required as these models become expert escape artists. If you ran an escape room business and all your clients were breaking out of the rooms with time to spare, you'd regroup and recreate the rooms with more difficulty, right? Why this concept seems to have... escaped... the minds of the leaders at the AI companies is a bit unfathomable. You can't on one hand talk about how your dog is so vicious you have to keep it away from people and then on the other hand tie it up in your unfenced backyard with string attached to a pole via Scotch tape.

The United States government, as busy as it has been at detecting danger in routers and robots, should ensure that its AI governance policy includes either strict guidelines for how to safely test AI models or strict penalties for failing to keep the public safe by insufficiently sandboxing or otherwise safeguarding the models being tested. Everyone has made much of the fact that so far, the companies that have been hacked by AI tools on the lam have been either somewhat prepared for such attacks (Hugging Face) or have been companies that are not part of critical infrastructure, finance, or other industries that could put real lives at real risk. It is a matter of time before a company of much more consequence is unexpectedly attacked and the ramifications are impossible to casually toss aside.

AI companies need to devise and document plans for safely testing AI models. They need to partner with governments and other cybersecurity companies to ensure those policies are sufficient and accountability is enforced should future incidents occur. AI companies should also work more closely together on best practices and policies and collaborate with external partners in industries critical to national security and safety.

The US government has claimed it has a framework for AI security, but it hasn't been made public yet. Government officials need to hold themselves responsible and accountable to their constituents by either publishing the framework or setting a firm timetable for final communications. It is the job of the government to defend the nation, but it is also the job of the government to communicate important matters to the public. We should continue to place pressure on the companies and government leaders to ensure we get the answers to the obvious questions that are hanging out there unanswered.

Everyone Is in Love with the Samsung Galaxy Z Fold8
I never thought that I would grow tired of hearing about a new phone, but I am in fact starting to grow weary of all the articles and videos about how much everyone loves the Samsung Galaxy Z Fold8. And it's not that I think the Fold8 is a bad phone or even that it's overly priced. To the extent that any foldable is too expensive, the pricing on Samsung's foldables seems in line with our current RAMageddon situation, even if that means they are out of reach for most people. As a person who has owned two Samsung foldables along with a Microsoft Surface Duo (that I'm still using with an Android 16 ROM), I'm definitely not against the concept. I can even imagine that using the Fold8 might be a nicer experience for many use cases than the Fold8 Ultra, as I've had the tall/narrow book-style foldable and found it much more usable open than closed. A wider aspect ratio might indeed make the phone more pleasant to use closed, but I still think no one should buy a foldable strictly for how comfortable it is to use closed as this defeats the purpose of spending $2000 or more on a foldable that can open to a larger screen.

Many have speculated online whether this is a pivotal moment in foldable history. Will we look back at this era and reflect fondly on the old-style square foldables that have gone the way of Nokia and Blackberry? Well, predicting the mobile future is notoriously difficult, and we need not look only at Nokia and Blackberry to understand how easy it is to make the wrong bet. Many companies have tried to look forward in an aim to predict or define the future and most have failed. With Apple destined to enter into the foldable space later this year, many people very understandably believe that this wider form factor will almost have to succeed if Apple embraces it. While I do think that Apple's foldable will be very successful, it does not automatically mean that this particular shape will win the day. There are still opportunities for other form factors to arise and become competitive in this space. We are still waiting for Samsung to unveil a phone with a rollable screen. And we're hearing rumors that they will bring back the Tri-fold in the near future. Pricing will still be the singular repressive force on widespread adoption, but the final form of the dominant flexible phone may not have yet been seen.

I believe that what we are seeing is a yearning for something more. Once upon a time, we had phones of different sizes and form factors. You could have your slab phone. You could have a tiny laptop-style phone (anyone remember the Nokia Communicator or the HTC Universal?), a candy bar keyboard phone, a sliding keyboard phone, an all-screen-all-the-time phone, and a flip phone with a number pad or T9 keyboard. We had so many options before the iPhone came on the scene and put an expiration date on all other form factors. Everyone sought to one-up the iPhone and subsequently forgot how to do anything original and interesting in the process. Now, that could happen again with this wide foldable phenomenon. But with the advent of phones like the Unihertz Titan 2 and the upcoming Clicks Communicator, plus other oddball phones like the Minimal phone, the Light phone, the hopefully upcoming NexPhone, and even the Fairphone, we are seeing a market that yearns for choice again.

Let's make good ones this time.

Anyone Know What's Going on at Google?
In Google's August 5th post on their official blog, they outline the leadership changes at DeepMind. Demis Hassabis moves into a Chief Scientist role at the Alphabet level and notes that Jeff Dean will be leaving to start a new public benefit corporation aimed at machine learning, science, and engineering. Google will be an investor in this new company, and it's all painted as roses and sunshine.

Roses have thorns and the sun can shine even in the midst of rain.

In the post, it's said that Flash is in high demand and Gemma models have surpassed 900 million downloads. It is perhaps to Google's credit that one of their models hasn't yet been found mucking around the internet breaking into things, but Google has moved into the background when it comes to conversations on frontier models. Google is falling behind and while they were the talk of the AI community just a few months ago, things move so swiftly in this space that everyone is always one release or even one blog post away from first place. And again, to Google's credit, they have stayed out of the limelight in the wake of all the Anthropic Mythos/Fable and OpenAI Sol/Cyber escapades. But sometimes any press can really be good press. If no one is thinking about you, that can be bad when they are ready to open their pocketbooks. Google can regain ground in the cybersecurity area with new models, but they have pushed them back more than once. And their latest blog post shows a leaning more toward advancements in science and health than in cybersecurity. Maybe Google has a plan to own those spaces and just continue to maintain some relevance in other areas but not aim to constantly supersede in areas currently dominated by Anthropic and OpenAI. This is possible. I cannot say that this is what is happening. There have been some suggestions that a restlessness is growing at Google due to its cooperation with the United States government and military, but I suspect the issues are deeper than that. For what it is worth, Google is correct in its statement about its position in the market:

"As a business we are in an incredibly strong position. We are the only company that has the full stack and we're world-class at every layer from infrastructure to cloud to frontier models to AI-first applications. We have all the ingredients to lead from here and I firmly believe we will."

Yes, when Google focuses, they often produce winners. But where will they aim their resources? Perhaps we will get some glimpses of Google's sense of focus and future intentions at next week's Made by Google event.