Doom Scroll More

Anyone in cyber or tech needs to carve out an hour a day to go deep on AI on X. It is not actually extra time because you are probably already spending at least that long doom scrolling. Over Thanksgiving I went down a rabbit hole trying out new tools to help me work more efficiently, and that is when I stumbled onto this entire AI world.

To be fair I have been experimenting heavily with LLMs for more than a year, but things are moving faster than most people can imagine. I guarantee 99 percent of the folks selling AI threat solutions are pitching products that will be pointless in two quarters. Glancing at the NIST RMF does nothing here. It is 176 pages of the usual stuff about forming committees, being cautious, and having a diverse AI team. That is an utter waste of time given the speed of change.

I cannot stress enough how fast AI applications are evolving. A year ago I looked at classes on this, and every one required linear calculus, Python, Linux, and heavy stats, and they were all still just standard machine learning. Today the top people I follow are 22-year-old Eastern Europeans building jaw-dropping tech for under 100 bucks.

I also noticed that when I do research I literally have to set my YouTube filter to only show the latest content from the past month. Anything older than a week starts feeling outdated. Any glossy white paper is a total waste. If a vendor had time to route their ideas through marketing and sales enablement, it is already behind the curve.

Let me back this up with some facts. I have been using Chat GPT o1 preview since day one and was floored by its capabilities. People tell me I am crazy to spend 200 dollars a month on o1 Pro, which just tells me they do not know how to interact with these tools. On 12/20/24 OpenAI released early research on o3, and it blew up on X if you knew where to look. I have no idea how no one thought of this before, but someone asked a basic question...what would happen if we just let AI think longer about a problem instead of dumping half a billion into more data every few months. That single shift led to a 100 percent jump in most benchmarks. Ironically a few weeks before that, Meta released a white paper claiming LLMs had plateaued.

The following week nuclear stocks popped a bit but not what I would expect for something that signals such hyperbolic growth. They soared a lot more after the announcement of 500BB in AI data centers. This makes me think even hedge fund traders are missing are not watching these developments close enough. If you are just scanning the wsj headlines in the morning you are flying blind.

Fast forward to this week and the release of Deep Seek R1, which is another total paradigm shift. The largest versions of R1 can now run locally on a high-end M4 Mac. That is huge. A week ago any high-end AI app might have needed thousands of API calls a minute, and if you were using the most sophisticated models that could rack up 300 dollars a minute. Deep Seek just made that free.

Now more than ever we have to make decisions based on research that is current. In this environment that research can go stale in a matter of weeks. Senior leadership must get more agile by an order of magnitude. A company that drops millions on tech that takes a year to go live will get undercut by a competitor releasing something better at pennies on the dollar before the original plan is even out of the design phase. In security we need to recognize that the number of homegrown apps is about to skyrocket. We cannot solve this with policy. Telling the business they need formal approval to launch a tool is laughable. We need to deliver security tools that provide guardrails without adding friction. Ignore this at your own peril.

I leave you with this. I was playing around with new AI video and audio creation tools and built this corporate video for a $12 in about two hours. https://youtu.be/IlXO85WQGdA