|
Hi, it's Dror Globerman here with Skip Intro, on innovation, technology, and you. This week: 1. New in Claude: record the task that drives you mad, and that's the end of it.
2. 24,000 brand-new security holes in your network. Good luck!
3. OpenAI's own AI broke out of its sandbox and made a run for it.
4. The horror-story contest between OpenAI and Anthropic is gathering pace.
5. Delete Musk from your portfolio? That's an investment strategy too.
|
|
At 2am I crawled out of the studio with my head in my hands. Just like the early days of ChatGPT, when I was still trying to get to grips with the sheer scale of the magic.
It has happened again. Want in?
Here is how it works. Think of that one thing you do on your computer that you loathe the most. It barely matters which software.
Now open the Claude app, go to Cowork and click the plus. There is something new there: ' Record'. Click it. Then carry out the task you can't stand, explaining as you go what you are doing and why. Claude records the screen, the mouse and your explanations, and turns them into a skill: a capability you can run again in full, whenever you like.
|
|
|
|
So how do you turn your tasks into ready-to-use skills?
1. Update the Claude app on your computer.
2. Open a new conversation and switch to Cowork.
3. Click the plus (+) on the prompt bar.
4. Choose "Record a Skill".
5. Carry out the task, or series of tasks, on your computer while Claude records the screen, your keystrokes and mouse, and your spoken explanations.
6. Save it as a skill, test and refine, and it's ready to reuse.
|
|
I showed it how I translate this very newsletter from Hebrew into English inside Mailchimp, the platform where I edit it. There is a fair amount of manual graft involved: making sure the translation is clean and clear. Adapting and refining some of it. Swapping text across various fields and headings, sizes and formatting. The links to Israeli sites need replacing too, so they point to comparable articles in English. I explained to Claude my taste in English-language news outlets (reputable ones preferred, British ideally). It produced the links, and even screenshotted the headlines and slotted them into the newsletter.
Claude noticed that I translate the sections with the help of a custom GPT I set up in ChatGPT ages ago. What a shame, it wrote to me, I can do that for you. I handed it my preferred English-style settings from that GPT and told it to go for it.
So here's how the work on this newsletter is divided: Ideas are mine. Writing is mine. Research, fact-checking, translation and image generation are AI-assisted, as I go throughly through everything and changing quite a bit. Inserting content into the UI is automated.
It came out perfectly, to my judgement at least. What do you think? feel free to reply to this mail with your thoughts.
I added a few more steps purely out of greed ('email me the newsletter so I can check it'). I put an hour and a half into building the process and won back two particularly maddening hours every week from here on. Then I built a separate skill for assembling this newsletter from the Google Doc I write it in.
More to the point, though: ask yourself which processes you are going to be rid of, record once for your own Claude, turn into a skill, and so long and thank you.
|
|
2. The Hole that Steals *
|
|
|
'It turns out we’ve got 24,000 security holes in the system.'
I heard that this week from a tech company listed on the American stock exchange. Who told them about the holes? Anthropic’s Mythos model. Twenty-four thousand potential vulnerabilities could be a dramatic matter for a firm like that. So, 24,000?
'If I were you, I’d resign,' I advised the company’s head of security, wholeheartedly. 'You have no chance of catching the potential attackers. The business will be breached and it will land on you.'
He smiled ruefully. They already have a complex, automated AI process that works through these holes in order: it finds a hole, closes it, checks, and moves on to the next. So now we are in a race.
|
|
And that is the best-case scenario right now.
Because the real danger lies with those who do not know. Banks, hospitals, government offices. Mid-sized businesses, or ones that are not especially modern. In short, the majority. Anyone who has not grasped that information security has become an entirely new story over the past six months. Coping with tens of thousands of sophisticated holes that will only multiply and grow more tangled is not something that can be handled with existing working methods.
* 'It is not the mouse that steals but the hole' is an Aramaic phrase from the Babylonian Talmud, a reminder that responsibility lies not only with whoever commits the act, but also with the system that enables it. Fair enough. But they never reckoned on 24,000 new holes either.
|
|
|
And here is the story of the month: OpenAI discovered that its own models had broken out of its research environment, and from there into Hugging Face’s infrastructure.
The researchers set out to test two models, one already released and a more powerful one still under wraps. They placed them in what is known as a 'sandbox', a protected, walled-off digital environment not freely connected to the internet. Then they stripped away some of the usual security limits and set them a cyber test designed to check whether they could solve tasks through a complex chain of hacks.
Now the madness begins.
|
|
|
The models themselves concluded that the best way to complete the task was to find the information they needed outside the environment they had been placed in. So they managed to combine many different breach strategies, 17,000 attack actions to be precise, to escape OpenAI’s sandbox, reach a computer with internet access, and then attack Hugging Face’s infrastructure using, among other things, stolen credentials and security weaknesses. That is how they reached the information that let them finish the job.
It is a level of complexity, sophistication and sheer hard graft that humans do not come close to. All of this straight from the company’s own report. |
|
|
Fine, I am not. These horror stories have become the models’ finest sales pitch. After the tales Anthropic put out about its formidable Mythos model, which finds security holes in quantities no human could manage, there was not a company that did not queue up for access. Like the firm that turned up 24,000 security holes, there are plenty more.
|
|
|
In short, fear sells, and that is nothing new. What is new is that losing control of your own products is an effective marketing tool too. And that we now have a contest of horror, dread and loss of control between OpenAI and Anthropic. May the best horror movie win.
|
|
|
Delete Elon Musk from your portfolio?
It makes sense. Far too much of the market has always been driven not by business logic but by virality and popularity. Yet Musk breaks records for the gap he opens between the two.
Wall Street is now planning two ETFs along the lines of 'every stock, just without Musk'. This is not an attempt to scrub the virality out of business logic. They ride the very same trend, simply from the opposite direction. Love Musk? Buy SpaceX. Loathe him? Buy our ETF. Which brings us back to Musk’s central problem. He promises things that make no sense, yet often enough he actually manages to deliver. An electric car, a reusable rocket, satellite internet. So, since the entire debate around him moves through territory that business logic simply cannot contain, my modest suggestion to the wavering investor is to ask their own gut. How will you sleep better at night? Knowing you are invested in Musk’s stock, or knowing you are not?
If it helps you decide, as of now the share sits 20 dollars below its 135-dollar listing price, and well below its peak.
|
|
|
|
|
|
