March 5th, 2030
It’s an unusually warm morning here in Austin. I ponder laying in bed for another 20 minutes, but it’s almost 8 and my wife has already started her morning yoga session. I groggily get up and head to the coffee maker which should have detected me being awake via my smart watch and started making coffee. We’re trying to keep our home relatively IoT free given issues with obsolescence and security, but a smart coffee maker is one of the few luxuries I’ve indulged in.
“It’s weird how IoT never became ubiquitous as we had assumed 15 years ago” I say absent mindedly, mostly to myself, but also to Kitsu, our helper robot, who is currently staring into space. We’ve had it for a few months, and besides its habit to just stare at nothing in particular for when it doesn’t have anything to do, it has been a real boon to our household. Does it count as IoT? Have I given up my principles? I briefly ponder this moral conundrum while I think about the robotics industry and its recent expansion. There’s been a few models that came out in the past 2-3 years of varying quality and capability, but a new European start-up launched this model recently at just over 5 thousand dollars, a price which I found acceptable given their strong focus on safety alignment and privacy. I think they could have made it a bit more human and approachable in its mannerism, but those Europeans have always been a colder bunch.
I grab my coffee and look around the kitchen. The dishes have been cleaned and arranged in their place. The fridge, microwave and counter tops are also spotless. Kitsu cleans them quietly overnight or when we’re not around, as I’ve instructed it to do. I hate the sound of a vacuum, so it also only vacuums when we leave the house. Laundry has also been done and the clothes are folded neatly in the closet.
I was skeptical about robots, but Kitsu has been a real help. It’s not very tall, just over 5 feet. It runs an onboard LLM engine which it uses to plan things around. According to the startup’s documentation, it’s able to look at a task (such as cleaning dishes), break it down into components and then write small routines and internal programs to achieve those tasks. Inference speed has increased tremendously but it can still be a bit too slow for real time actions and responses and generating software doesn’t cost anything nowadays, so this seems like a valid compromise. I know it can connect to the cloud to access a better LLM, but I’ve found it perfectly capable just running the onboard one.
One of my buddies, Joe, bought like 4 of those things and some more from other brands. He has decided to renovate his house purely using robots, and the results haven’t been half bad. His robo-gang still needs a lot of supervision and direction, but his new kitchen is shaping up nicely. There’s some mesh technology that helps them synchronize so they’re pretty good at dividing tasks.
After a quick walk, I get to my home office room. It’s not quite 9 o’clock yet, so instead of sitting down at my work laptop, I check my personal one. I’ve had some new ideas on how to obfuscate payloads in malware, and I’ve tasked one of the many LLM agents I use to build me a lab and a pipeline to test this new implementation. Last night, before getting into bed, I tried out my idea and my payload still got picked up by EDR. I tasked the LLM agent to explore my idea and come up with some alternative ways to implement them and run those experiments by itself. Now I’m glancing over the results and seeing a lot of iterative approaches, but I finally see it! Attempt 36 got past detection. I painfully recall how this kind of experimentation used to take me weeks, or even months, just 7-8 years ago and I am glad that agentic LLMs are everywhere now. I make a mental note of the result and prepare to write a blog post later on my discovery. Writing is still something I do manually, in part because it’s enjoyable, but mainly because through writing I get to really understand and explore an idea. I still haven’t seen any LLM write something like the Cryptonomicon, but they’re excellent writers and listeners in a corporate setting.
It’s 9 so I log onto my work laptop. I check the result of the mirriad of agents that we have running that handle scoping, reconnaissance, pentesting, remediation, reporting and communication. It looks like 3 more assessments have been completed this week, but one of them needs my attention. I ask the agent to bring me up to speed with the architecture of the webapp and the existing findings and where it got stuck. I see it, a possible Server-Side Request Forgery (SSRF) vector. LLMs still have trouble with this one, so I go through the things it has already tried. It seems like the LLM agent ended up going a bit in circles, trying OS commands and URLs. I make some educated guesses based on the rest of the architecture and try to hit port 10443 on some arbitrary internal hosts. I noticed a few of them behaved differently. The LLM agent quickly steps in and mentions something about fortinet and that it sees the way forward now. I let it do its thing while I go back to admiring our agentic pipeline.
It took some work to set it up, but now with 5 people we are doing the work of 30. We still have open roles for our team, but we are looking only for senior people who would be able to make a sizable impact on our pipeline. We haven’t hired a junior in 3 years. Wouldn’t want to be a fresh graduate now. Unemployment keeps rising steadily and it’s at almost 10% now. Some people are tense but there haven’t been major political moves to address it yet, just a lot of posturing.
I kick my feet up and look around. It’s an odd feeling. A lot has changed in the past 5 years, but some things haven’t. My car finally drives itself 99% of the time. I have a robot in the house. I get to enjoy my work and hobbies because LLMs handle the grunt work. I even have an LLM fitness coach now that can monitor my nutrition and form via my phone camera, although we still handle most of the cooking ourselves. Turns out cooking requires a lot of fast thinking, dexterity and adaptability, which Kitsu still struggles with. It does make some great snacks though!
On the other hand, LLMs seem to have hit a bit of a plateau. GPT 6 came out recently and everyone says AGI is here. It is certainly smarter than 99% of the people in 99% of the cases, in the same way a calculator can do maths faster than any human. Because of this, it still needs a human in the loop most of the time. A whole new industry and way of thinking has sprung up centered just on making agents more efficient. Short, tight loops with a lot of reasoning has been a real breakthrough in improving quality.
Scientists are now using LLM agents to research other ways to achieve “real AGI” or ASI. I used to be quite into the race to AGI movement, but the goalposts seem to be arbitrarily moved around whenever companies need to raise money again. There doesn’t seem to have been a lot of progress. Space travel and trips are still out of reach, but things seem to be progressing faster. People still spend too much time on social media loving and hating (mostly hating) each other. I think they’re still people? I used to be able to tell the LLM responses apart from the genuine ones but not so much anymore. It’s sort of incredible that we have access to such a level of intelligence but it hasn’t really solved any of our fundamental issues. There’s still war. There’s still sickness. I read news all the time about how medicine development and research is faster than ever, but it seems like the really difficult problems like cancer and aging haven’t been cracked yet. Perhaps LLMs have some fundamental limitations in their thinking. Much like a plane doesn’t fly like a bird but achieves a similar result. Or perhaps GPT 7 will solve everything. I don’t know.
I get up to get my 2nd cup of coffee. I pass by my wife and ask her if she wants to watch that new AI generated anime that’s all over the internet later on. She scoffs at the idea and says something about it being creatively bankrupt. I guess artists have more principles than us tech folk. We agree to keep watching JoJo. I get back to my desk and start writing a blog post. I think this one will be about today, March 5th, 2030.