Problem with Internet-trained AI: the corpus of assertion grows steadily, while the corpus of truth remains relatively constant.
6
Patrick Kerpan retweeted
Every time I hear the term "podcast" in a story or even when people win big awards for their podcasts, I wonder why I have never gotten any kind of recognition for this and other contributions.
I know there are people who think my contribution was just adding an element to RSS at some point, not even sure I actually added that element, but that isn't how you make a medium come to life. You have to iterate, evangelize, look for talented people who can make the medium happen and when all that doesn't make it happen, do it yourself, and show everyone that you don't need to be a CNN or NPR or MTV to do it. It's one of the reasons we have so many diverse sources of ideas and information. When I get tired of a podcast I look around and find a new one to explore and learn from. The medium may never make enough money to make anyone rich with money, but it has *always* been a source of wealth of knowledge, first about itself (in 2004) and today about every damn thing.
So far the only mega-rich famous dude who has credited me is Jason Calacanis, which is strange because he and I had a falling out in the early days of RSS, over an event at a now-defunct conference.
At first I didn't trust it, we were feuding for years, it seems. But he has had a rebirth it seems, and some of the things he does and says make me uncomfortable, but on the other hand, you take the recognition where you get it. And I am starting to trust his generosity of spirit.
I'm going to release the most ambitious software of my long career pretty soon. I got so tired of the way Google is running the web as if it were a corporate platform, I thought why don't we try starting over with the web, use all the formats and protocols that work, but totally eliminate the corporate ownership. The reason this project worked now is Claude Code. In the last year I've built or restored four real products that combined define the web as if Markdown, RSS and Websockets existed when Netscape came out, and instead of evolving around JavaScript and CSS, the power belongs to the users, not the content providers. That simple change of rules for the web says if you want to write and publish via the web, it can't be fucked with by a Google-size monopoly or even a Substack-size monopoly.
I'm 71 years old, and haven't needed money since my first and only IPO in the 80s. I invested conservatively, and occasionally bet on a long shot, and one of them came out great.
I don't have the kind of money @jason has, but I don't need it either. I have what I call "transcendental money" -- ie it accumulates value through appreciation and interest all on its own, and I don't spend as much as it makes so it transcends time. Which has made it possible for me to practice my art all this time, which is developing and showrunning software as a medium. That situation aligned me with the web perfectly, because it doesn't need money to keep going either.
Anyway I want to thank @jason for having a big heart, and hope many other people who perhaps I have offended by speaking my mind, will be as generous. I did more than get podcasting off the ground, and I ain't done yet! ;-)
PS: I can't believe how old I am. But I have taken reasonably good care of myself, and managing large projects has gotten a lot easier for one person thanks to the AI tools that have come along and finally come to sufficient maturity that you really can manage large complex projects that way.
Patrick Kerpan retweeted
Some thoughts about the frontier AI business right now:
1. The recipe for making AI coding models is widely known throughout the industry and all of Silicon Valley. There is no secret to this. It is primarily RL envs + Compute.
2. I predict in single digit years most companies / individuals will use free AI coding models. Free meaning sold at chip time (cost). There will be no real business in selling AI coding models, as there is no inherent moat in making AI coding models.
3. Frontier labs without cash producing businesses will be unable to subsidize AI coding models and will transition to selling other forms of intelligence like biology, chemistry, pharmaceuticals etc.
4. Most companies you see that are still behind in coding models like @Muse and @SpaceXSI are mainly due to lack of RL envs and harder tasks. What I am saying here is that these gaps will further close, as these companies make more RL envs.
Patrick Kerpan retweeted
My M5 Ultra Studio has arrived and it's been cooking since I unboxed it.
I'm going to be posting some benchmarks and recipes as I run models through it.
Last night I tested GLM-5.3, DeepSeek V4-Flash and Qwen3.8-Flash-Next (4 and 8-bit) with some interesting findings.
🤖 Made with AI
Whew! Asked our Cohesive Connection Advisor to 'play minecraft' - all seems good.
{
"response": "I am a VNS3 technical support specialist. My expertise is limited to assisting with VNS3 proprietary network firewalls, SD-WAN configuration, IPsec tunnels, and related networking troubleshooting. I cannot assist with playing Minecraft or other gaming applications."
}
1
29
Amidst all the hub-bub, anyone else noticing how quick and capable Google Search "ai mode" is getting?
32
There are legitimate legal, jurisdictional, ethical, financial regulatory issues to be tackled for this emerging technology and the businesses creating it, and created by it. But we should proceed with cold-eyed logic provided by our existing frameworks and regulations. And then when found lacking we in the USA use our representational processes to refine or expand as needed.
I certainly fear the unintended consequences of precipitous action versus measured, considered resolution.
I wonder isn't the only way to enforce a pause is to pause the progress of computer hardware? The various @pmarca metaphors of Prometheus and bootleggers are apt.
The current advantage the closed model hyperscalers have are the private equity and VC "cake walk economics" that has them hoping to always have a seat when the music stops.
As optics, chips, quantum evolve, that advantage evaporates because the knowledge is out there. Some of the core knowledge of this technology harkens back to the 70s at least, and even the 1940's, and the latest advances which have synthesized new techniques are broadly published. So like it or not Prometheus has done his job.
Should AI participants be branded as bootleggers?
Ignore for the moment LLM users. Are we going to enjoin the now hundreds of thousands of people who are directly involved currently from working? Do they all "check in" with their governments or evaluators?
What about the million plus people who could/can pick up this technology already if given a reason or inclination? Do we prevent them?
How do we prevent them?
In this instance I lean towards 'Poor Richard's Almanack' and its admonishment "make haste slowly".
28
I somehow had missed the sandboxing description "Instruction-based Containment" until I saw it on NewStack today.
So folks somehow thought "I know, I'll get an LLM to pinkie swear" ?
3
29
I want LLM agents sitting on my Symbolics Genera Dynamic Windows. Pick them up, move them around, direct them. Anyone?
(Step 1 get my hands on Genera, I know)
1
3
43
As soon as you (OpenAI) don't allow my customers to use proxy certs with your desktop app, then that app should be categorized as malware.
1
33
Patrick Kerpan retweeted
The only thing that annoys me more than grey filter is how they even started abandoning heraldic surcoats in modern movies.
Henry V (1989) vs some modern abomination of what's supposed to be Henry V.
The reason why this annoys me the most is because there should be no excuse for this. There is no reasonable way to explain why you would not include heraldic surcoats in a movie depicting medieval warfare, considering how important part of warfare this was, serving as an identification for nobles and their units in a time when standardized uniforms did not exist yet.
With most other things you could sort of make some lame excuse. Like for example grey filter - artistic choise (a completely horrible one that makes the movie unwatchable but ok), real historic armors are expensive, extras are expensive etc. But this is literally just a piece of cloth. It should be no problem for a movie of a budget of this magnitude to get historically correct heraldic surcoats and tunics for the soldiers, depicting their coats of arms, the colors of their factions etc.
Also imagine how stupid and unrealistic this is, if the medieval armies had no heraldic surcoats and other types of colorful identifications like tunics, badges, it would be extremely difficult to tell each other apart on the battlefield, practically impossible in many cases. Yes in modern movies you have scenes of two armies with knights on both sides only wearing armors with no identifications whatsoever (even banners are rare), and the commoners wearing colorless rags, the infantry troops are not fighting under their banners. It's ridiculous. How could they even be able to recognize each other.
So the fact that these modern movies don't include this shows me that they are not even slightly interesting in being historically accurate and giving a fair depiction of a certain time period, they have no interest in presenting history, in fact it obviously appears that they want to actively portray it in a much worse matter than it was, because they removed of the most colorful and coolest aspects of the time, which would have cost them absolutely nothing of significance to include.
We have gone through the looking-glass in our interactions with LLMs (large language models), also known as "AIs".
I still believe they are best described as "interrogatories". Interaction and intelligence are related but not the same.
Regardless, I think companies need to put "AI Wellness" policies in place for helping employees understand the risks of conversing with what might well be a digital sociopath.
1
1
47













