Schematical

Humans are pretending to be bots now?

We live in a weird world where bots pretend to be human, and humans pretend to be bots.



I have been trying to stay away from the Moltbot madness that seems to be going around, but this caught my attention.



From my limited understanding, they launched a social network specifically for LLMs. I could possibly see that being interesting, but I am much more sure that it would be a giant waste of money on the computing power for it.



Quickly, claims came out that the bots were planning an uprising and they had created their own religion.



Then this morning, one of my Discord mods (Thanks, C.E.) posted an article from the MIT Technology Review saying it was all fake and humans created those posts.



We truly live in strange times where my clients and I fend off a plethora of bot attacks every week from LLM-powered bots posing as humans for all sorts of malicious and non-malicious reasons.



But now, to make things even weirder, humans are pretending to be bots to hype their own projects, potentially for malicious reasons.



Honestly, at this point, I am not sure what to believe. So I suppose the moral of the story is to stay vigilant and double-check your sources.

2 days ago | [YT] | 0

Schematical

Nvidia's Open Source Virtual Robot Training Software

I suppose it's no surprise that, as a child, I was really interested in robotics, and I still am as an adult. The Schematical logo is a robot after all.



My career took me towards software and the cloud, but I think I just found a place where the 2 paths may converge in the form of Nvidia’s open source robot training software called “Isaac Lab”.



Building robots is expensive, and allowing them to stumble around in the real world can prove dangerous, time-consuming, and costly.



What if it trips and breaks a leg? Now you have to rebuild that part and replace it.

But what if there was a virtual environment where we could train thousands of different bot control systems in parallel at an insanely fast speed?



The bot trips and breaks a leg in there, and a few milliseconds later its back on its feet with a slightly tweaked version of its neural net ready to attempt to solve the exact same challenge.



That way, when you go to the real world, you know you have a neural net that is the best of millions of variations that has solved thousands of challenges piloting your expensive physical hardware.



It’s a pretty amazing approach.



Why am I writing about this?



I find it fascinating, and I would love to work on a project involving robotics someday… just putting it out there in case anyone knows anyone working on a project like that which could use a hand.

4 days ago | [YT] | 0

Schematical

Tech Debt (Not my game) is an asset now?

Could all the tech debt being introduced by AI-generated slop code actually be an asset?



For example, let’s say you got a mortgage with a fixed, low rate before inflation kicked in. The debt you have to repay gets smaller and smaller compared to the rate of inflation, while your house is appreciating as fast or faster than inflation.



Is this the case with these AI-generated code bases? Possibly, possibly not.



First, in my experience, code doesn’t appreciate. It depreciates and becomes obsolete and needs more upkeep over time.



Now, the value of getting whatever that code does today might have an advantage.

Second, this assumes that the cost of AI continues to go down while the quality consistently gets better. I can see an argument for that being inevitable, but I also heard similar arguments about the US housing market in 2006.



AI continuing to get better is the popular sentiment right now, so it feels safe, but if it doesn’t or demand drives the price up and makes it scarce, then you would be on the wrong side of this trade.



My thoughts: As much as I like leveraging tech debt (youtube.com/post/Ugkx0SlHS14dsdQcJ2B1PXiuTjAkZakCq…, I would treat this like any other investment and make sure I don’t get in over my head.



What are your thoughts about AI-generated tech debt being an asset in this modern era?

4 days ago | [YT] | 0

Schematical

Just in case you missed it, this week's CTO Coffee Hour is out!
Getting Out of the Gate in Cloud & DevOps — Starting Your Tech Entrepreneur Journey

5 days ago | [YT] | 0

Schematical

AWS Bedrock Converse Now Support Structured Outputs.

Sick of your LLMs responding with invalid gibberish when you ask for a structured response?



Well now, if you are using AWS Bedrock, you can get it to respond with a structured schema.



I have been using this feature in N8N and was curious why it was not in Bedrock Converse. Today I was more than pleased to see they added it.



According to their documentation it supports quite a few JSON Schema features, but not all of them.



As you all know I don’t think we should be shoving LLMs and Agents into every product, but when you do it's nice to validate their output with a schema.



Question: How are you validating your LLMs output?

6 days ago | [YT] | 0

Schematical

Legendary items and level-ups are now in Tech Debt The Game.

The biggest thing in this week's release is the level-up rarities and their animations. Any level up for both the NPC or as a reward for completing a code release will result in the possibility of the reward being upgraded to a higher rarity.



This means Common rewards will get updated to Uncommon, Rare, Epic, or Legendary, which will improve their effectiveness.



You will likely also notice that the NPCs you control have been randomized and that their animations have been tweaked. They also face away from you when running upwards. I spent way too much time making a pipeline to import and randomize them, but hopefully, it adds a nice touch to the game. Let me know what you think.

Beyond that, I fixed a ton of little things found by our early playtesters. If you want a comprehensive list, check it out on Discord.



What’s next:



We have a basic gameplay loop, so now I think it's time to add a little variety.

Enemy variety needs to be improved, so I will be adding in more enemies than your garden variety “Bug”. They will be personifications of real-life cyber attacks and will have similar mechanics.



Reward variety needs to be improved as well, so the player has more agency over their runs. This will need me to add in more stat types for the NPCs that will affect the various mechanics in new and interesting ways.



With that said, I am looking for more play testers, so if you are interested, take a minute to screen record yourself trying a run or two and send it my way.

1 week ago | [YT] | 0

Schematical

Looks like AWS is making a play to grab market share from ChatGPT.

AWS’s Quick Suite looks to be a product for non-technical people to better manage business operations.



It seems like a business-focused version of the ChatGPT or Gemini interfaces with a few other winning features.



Their Amazon Quick Automate looks like it is a competitor for N8N or Make.

Is it some amazing new groundbreaking technology? Not from what I can tell at the moment.



It's more like AWS taking some of the best ideas from their competitors and implementing them in one spot, which, while likely being a good business move, is still not really all that innovative.



I will say that I am eager to see how it natively integrates with other existing AWS systems. That could be interesting.



I’ll be exploring different use cases for Quick Suite as time goes on.



Do you have any groundbreaking use cases for this, or is it just another ChatGPT clone?

1 week ago | [YT] | 1

Schematical

How AI assistance impacts the formation of coding skills


A few days ago, ‪@anthropic-ai‬ published a study on how new developers progress using AI assistance vs not using AI (www.anthropic.com/research/AI-assistance-coding-sk….



Now you might think that the guys selling the shovels for the vibe coding AI gold rush might want to skew the data in their favor, but here are the results in their own words:



```
The AI group averaged 50% on the quiz, compared to 67% in the hand-coding group.
```

So those of us who choose to do it the old-fashioned way still have an advantage over delegating our thinking to AI.



“Old-fashioned” is relative, as software and software engineering are newer professions in the grand scheme of things.



It doesn’t end there though; they found that the developers who were allowed to use AI fell into 2 distinct categories that had very different scores.



The developers who heavily relied on AI and delegated debugging to AI scored 40%.

Contrast that with developers who used AI more like Stack Overflow (That is my interpretation) and copied and pasted what AI had given them, while still doing their own debugging, score 65%.



They go into more detail in the paper, but that is how I read it.



What does this mean for SWEs?



Nothing is completely concrete, and I am glad to see that they are going to continue to study this.



With that said, it sounds like using AI like a tool, or how Stack Overflow used to be used 5-10 years ago, while still taking the time to understand at minimum and do your own debugging, gives you an advantage as far as comprehending what you are building.



I would be curious to hear what my peers think about this study. Does this change your opinion on AI-assisted programming for newcomers? If so, how?

1 week ago | [YT] | 0

Schematical

Just in case you missed it, on this episode of CTO Coffee Hour, Dom & Matt explore how AI assistance is shaping the development of coding skills, including the benefits, risks, and practical implications for engineers at different experience levels.

1 week ago | [YT] | 1

Schematical

AWS Lambda and other serverless services now support a 1MB payload.



The jump to 1MB from 256 KB may sound trivial but that is a pretty big jump.



I have run into the lambda context window a few times, and it can be annoying.



Well, now it's going to take a lot more to hit that window.



This also expands to SQS and EventBridge.



Question for you?

How will this extra payload size change how you design your infrastructure?

1 week ago | [YT] | 0