We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.
block.xyz/dessa
Been working on a learning tool project called Luca. Named after Luca Pacioli from 15th century Italy. More context I will share in a later post perhaps.
there’s this concept of 4 rungs there.
Orient aka Scout,
Apply aka Wield,
Rebuild aka Reforge,
Own aka Vouch.
this own aka vouch rung could be where the value lies in SWE at least for a while.
Why?
Still at the end of the day ownership and accountability lie with a human ( NI - natural intelligence agent ).
Why?
cause in principle for accountability to work it has to lie on someone that can feel loss. The loss from a bad call, at the end of the chain, still has to land on a named human who signed off.
Ok, so to own aka vouch for a system, you need to understand it fairly well. AI made building instant. It didn’t make understanding instant. Can use AI to make understanding faster. but not as instant as “building” due to the nature of what 'understanding' is.
Getting oriented in a codebase an AI wrote still takes time. making a change through every layer yourself takes more.
Vouching, being able to say “I understand this system, and if it breaks, I own it”, takes the most.
So this takes time and the tacit knowledge required to ask right questions. both of these are scarce. As per the basic rule of ecnomics scarcity combined with active demand creates value.
Not everyone can put in the time to be the person that vouches for everything. that puts their ass on the line.
so yeah this could be the place, SWE value lies at least for a while.
24
It's interesting to view human organizations as just another 'technology'. Or at least it is to me.
actually not just another technology. it's kind of THE technology that allowed us to do stuff our single monkey brains couldn't.
Anyhow not going to go deep into that. just going to share the below insightful talk by Adam Bender.
This talks about how Software engineering organizations are a complex adaptive system that is intertwined with it's enclosing environment ( engineering culture etc. ). And about viewing AI as an amplifier ( a thing with magnitude only, no direction ) that is now amplifying both the quote unquote 'good' and the 'bad' of these organizations.
Unless your system has measures to handle that, AI amplifies confusion just as much as the code generation it's amplifying.
As i mentioned at the start SWE orgs are a 'technology'. A system.
A web backend system with good strong foundations can scale up, when the traffic 10xs or 100xs. But a backend system with weak foundations will crumble under amplified traffic. Or simply won't be able to monetize that traffic increase.
That same analogy applies to software orgs.
This might sound intuitive. But it's surprising to see how many organizations neglect this fact.
( well on second thought it's not suprising actually when you think about it. )
Anyhow this talk is a must watch for anyone who's thinking how to patch their "organizations" on the fly, since most of us don't get the luxury of a greenfield rebuild.
youtu.be/2n41YjR5QfU
one uncomfortable open question i don't think this video covers.
a lot of our current process (code review, KTs) exists to coordinate multiple humans. When we're designing systems to leverage AI amplification, are we still designing around that coordination complexity? well if you don't, what about the excess labor? Sure, jevon's paradox might kick in and you could extend the scope and do things/projects that wasn't possible before. But what about the excess labor till then?
63
Software Engineering, the trillion dollar industry, is just all about ‘shaping’ text and ‘placing’ text.
Everything from building an agent native arch linux distro to porting a .NET service to Golang to deploying code from your editor into AWS lambda via a monorepo others are continuously pushing code is just shaping and placing text.
now bear with this reductionist shower thought of mine.
Shaping text - whether it's in the shape of 'assembly' instructions, 'c' procedures, 'c++' classes, 'javascript' objects, a compiled binary(go module gets shaped into binary once u run go build), OR whether it's in the shape of a functional, object oriented, multi paradigm dirty vibe coded ts code, tightly coupled, loosely coupled whatever it is, all of those are different ways of shaping the same thing into different kinds of text.
the decision on how to shape the text depends on factors like,
human cognitive limits. so optimize the "shape" for readability ( ex:- c over assembly, go over C)
performance ( ex:- rust over python )
adaptability and scalability ( ex:- clean architecture, high cohesion )
Placing text - placing text from your machine into a machine/s that's running 24/7. Also if you're writing this “text” for a bigger and greater purpose, ( like writing software for self driving cars or fixing a weird timezone bug in a legacy ruby on rails app at work ) , you can't do it alone ( or "couldn't" is the better verb given the synthetic intelligence future ). Why not alone? cause your monkey brain has cognitive limits. so you have to work with other monkeys who also want their text placed on the same machine/s. Plus the text you place might have to perform actions in other user "machine" via another piece of text called "web browser" or "mobile os". and all of this "placing text", business requries fancy terms like,
devops, agile, Jira, aws, kubernetes, platform eng, finops, division of labor etc.
Ok, so this distilled view is helpful and interesting when looking at SWE as a whole. anyhow let me also talk beyond the reduction. We humans have been tool builders aka engineers at our core since the day we made that pointy stone to stab that Hyena.
and SWE has been our way of just building more tools by “shaping“ text that can be quote unquote understood by this machine called computer. also now with our newest tool 'synthetic intelligence', everybody's thinking and talking how SWE will change.
I prefer to see a world where humans continue to operate in much higher levels than what was possible before, using 'synthetic intelligence' agents to enhance our cognitive capacities. So as i said everybody is putting out their opinions now. Some with more conviction than they should have I think. Anyhow, nobody can say for sure. After all, we're just a bit advanced monkeys trying to bite more than we can chew by engineering a bigger mouth. and hey, that could work. Just like we fly even though we don't have wings.
2
2
498
A.kal retweeted
I can’t believe Google dropped a 64-page guide on building AI Agents. Completely for free!
From AgentOps to Agentic RAG orchestration to grounding, this guide shows how to design and scale AI agents reliably in production ✨
If you’re building an AI startup, start here.
A.kal retweeted
🔑 Stolen private keys are the number 3 attack vector for DeFi for 2023.
The. Number. Three.
With over ~$137M stolen so far.
Sometimes, it's as simple as accidentally pushing private keys up to GitHub
Fret no more ripped Jesus enthusiasts, you have a solution now
👇
1/ How do you store your users' passwords? 🔐Not in plaintext, I hope. You're probably salting the pwd -> hash it -> store hash + salt. But is this the safest way? 🤔Hang on for a bit to find out.
#CyberSecurity
1
89
6/ I've explored this in-depth in my latest article - How to never store or even know your user’s passwords with SRP protocol and AWS Cognito?
1
63
7/ Curious to find out how this protocol works behind the scenes and how you can implement it effortlessly? Go check it out here. 👇🏼
akalankapathirage.medium.com…
45
4/ But, What if I told you that there's a way to authenticate users without ever accessing or even knowing their passwords? In other words what if u can have zero knowledge about your user's passwords? 0️⃣
#zkproofs
1
39
2/ Well, there have been numerous password leaks of well-known brands with millions of users in recent history proving this isn’t the most secure way.
1
65
1/ How do you store your users' passwords? 🔐Not in plaintext, I hope. You're probably salting the pwd -> hash it -> store hash + salt. But is this the safest way? 🤔Hang on for a bit to find out.
#CyberSecurity
A.kal retweeted
AI Twitter is flooded with low-quality stuff recently. No, GPT is not “dethroned”. And thin wrapper apps are not “insane”. At all.
I feel obligated to surface some quality posts I bookmarked. Every one of them should've been promoted 10x, but ¯\_(ツ)_/¯
In no particular order:
1/ 🚨 Smart contracts are not self-executing. Yeah, you heard me right. 👀
Read this thread to learn why is that and more.
#smartcontracts #blockchain #chainlink
13/ 👉When you send a transaction scheduler message to the Massa network, it will go to a pool called the async pool. You have to pay the gas cost upfront. At the moment Messa operates in testnet only. So this is currently at an experimental stage.
1
1
117












