What if the biggest challenge facing AI isnât innovation, but legitimacy?
As generative AI accelerates, concerns are moving beyond competition and copyright.
Amnesty International has raised a major warning about the foundations of generative AI.
In @amnesty new report, Unlawful by Design, the organization argues that many leading AI models may be built on large-scale privacy violations through web scraping practices that collect personal data without meaningful consent.
The report examines systems behind tools from Google, OpenAI, Meta, DeepSeek, Midjourney, and Stability AI.
But the concern goes beyond privacy.
Amnesty says these systems also raise broader human rights risks:
âą Amplifying bias and discrimination
âą Threatening freedom of thought
âą Increasing environmental strain through energy-intensive infrastructure
This shifts the AI debate into new territory.
For years, the focus has largely been on copyright and competition.
Amnesty reframes the issue around something bigger:
Human rights.
As AI becomes core digital infrastructure, the real challenge is not just building more powerful systems.
Itâs building systems that are transparent, accountable, and aligned with fundamental rights.
The future of AI will be shaped not only by what it can do, but by the principles it is built on.
Just built an insane new agent skill.
It can perfectly extract slides from YT videos, then write notes, images, transcripts, and slides into Obsidian vaults.
An HTML artifact allows me to navigate and add more notes as I listen.
Should I release the skill?
/goal is really insane!
It's how you can get the most out of coding agents today.
For efficiency, I find it works best when you do planning before /goal. This ensures the agent has the right context and goal, which often only happens with careful planning.
New research from Microsoft Research
I see a lot of AI engineers handwriting agent skill docs and hope they generalize.
Probably not optimal. This works show why.
It treats the skill doc as a trainable external state of a frozen agent instead.
It introduces SkillOpt, where an optimizer model makes validation-gated edits to the skill file. It adds, deletes, or replaces instructions, with a textual learning rate that controls how aggressively each round rewrites the doc. The agent itself never changes.
SkillOpt is best or tied on all 52 (model, benchmark, harness) cells.
On GPT-5.5 it adds 23.5 points in direct chat, 24.8 with Codex, and 19.1 with Claude Code over no skill. It beats human-written skills, TextGrad, GEPA, and EvoSkill, carries zero extra inference-time cost, and the learned skills transfer across models and harnesses.
Paper: https://t.co/mNgTmmT32U
Learn to build effective AI agents in our academy: https://t.co/1e8RZKs4uX
Most people still think building AI products requires:
â expensive APIs
â huge cloud bills
â VC funding
â a team of ML engineers
That was true in 2023.
In 2026, a single developer can build production-grade AI systems for basically $0.
Hereâs the modern AI stack nobody talks about enough đ
đ§ LLM Layer
Run models locally for free:
â Ollama
â Gemma 4
â Llama 3.3
â Mistral Small 4
No API costs.
No rate limits.
No sending private data to external providers.
âïž Agent Orchestration
Coordinate multi-step AI workflows with:
â LangGraph
â CrewAI
Agents can now:
âą research
âą reason
âą use tools
âą delegate tasks
âą validate outputs
This is where AI shifts from âchatbotâ â âsystem.â
đ RAG Stack
Production retrieval without paying OpenAI:
â LlamaIndex
â ChromaDB
â Qdrant
Build:
âą internal copilots
âą document search
âą memory systems
âą AI knowledge bases
âą autonomous research agents
Entirely local if you want.
đ Tool Layer
MCP changed everything.
Itâs becoming the universal protocol between:
â agents
â IDEs
â databases
â browsers
â APIs
â developer tools
One standard interface for AI systems to interact with the real world.
đ» Coding Agents
Your new AI engineering team:
â Claude Code CLI
â Aider
People are shipping entire startups with AI-assisted development loops.
The best developers now orchestrate agents instead of writing every line manually.
đ Frontend Layer
Ship interfaces instantly with:
â Next.js
â Streamlit
â Vercel free tier
You can go from idea â deployed AI app in a single day.
đïž Data Layer
Simple > overengineered:
â SQLite
â DuckDB
â Supabase free tier
Most AI apps do NOT need complex distributed infrastructure early on.
đ Observability
Serious AI systems need tracing:
â Langfuse
â Phoenix
Track:
âą hallucinations
âą latency
âą agent decisions
âą prompt failures
âą retrieval quality
Production AI without observability is blindfolded engineering.
đ Deployment
Deploy almost anywhere:
â Docker
â Cloudflare Workers
â HuggingFace Spaces
Global AI apps can now run on infrastructure cheaper than a Netflix subscription.
Total stack cost:
$0
The tools are free.
The real moat is:
â architecture knowledge
â workflow design
â context engineering
â system thinking
Not access to models.
Weâve entered the era where small teams can build systems that used to require entire companies.
Save this for your next AI build đ
#AIArchitecture #AgenticAI #LLM #BuildInPublic
Now pitching is Hayat Outahar from @GoodHiveLabs đ
A decentralized recruitment protocol for Web3 tech talentâusing token incentives and on-chain reputation to align hiring with community values.
#EthVC
In tech, we often talk about first-mover advantage. But increasingly, the real mass-market power lies with the second mover: the one who refines, scales, and distributes after the groundwork has been laid.
@reidhoffman puts it best:
âYou donât have to be first, you just have to be right.â
Look at @brexHQ vs. @tryramp.
@samdblond and the team at Brex spent 4 years building a GTM engine from scratch:
âąEducating the market
âąSponsoring conferences
âąPlastering billboards
âąSending custom gift boxes to fundraising CEOs
By the time Ramp arrived, customer education was done.
Ramp just had to copy the playbook, ship a smoother product, and win.
Weâre starting to see similar patterns in crypto:
âąSolana â Sei
âąBase â Robinhood
Solana and Base pioneered the category - investing deeply in infra, ecosystem building, and narrative shaping.
Now, Sei and Hood are showing signs of being second movers with cleaner UX, refined positioning, and the ability to scale downstream, without the same CAC burden.
Customer education is expensive.
But for second movers, timing is a weapon.
So pumped for this 25-28 consumer supercycle fam đ
We're launching the first-ever zkVerified markets on @avax with @eulerfinance@multiplifi@PythNetwork
This changes everything.
Institutional capital can finally access DeFi without sacrificing privacy or decentralization.
The man who can predict the future:
Balaji.
He was an early investor in Bitcoin, Ethereum and Eight Sleep.
He just said, "Social media is about to become far, far more lucrative than people realize."
His 4 shocking predictions on the future of social media & wealth creation: