Meet our team of AI & ML professionals from Belgrade! We partner with businesses to turn complex challenges into powerful, scalable MVPs & POCs, custom-built for each client.
Ready to build? Let's talk. ▶️ https://t.co/VxM56oIR6C
#AI#MachineLearning#WebDev#ProductDevelopment
Job&Talent put 4 AI agents to work: hiring, attendance, performance, support.
Result: 30%+ lower operating costs, 5M interviews at 80% lower cost, 70% of code AI-written.
Agents aren’t a pilot anymore. They’re the org chart. #AI#AIAgents
Harvey’s margins went negative on frontier models — it switched. AT&T moved 40% of AI work to open models.
The fix isn’t less AI. It’s routing: ~80% of requests don’t need a frontier model.
Your AI bill is an architecture problem.
#AI#AIEconomics
Two months after hitting 1B users, Gemini's free tier dropped from 3 models to 1.
AI Plus loses Pro — same price, less model. The $19.99 tier gets the carrot.
First you win the billion. Then you invoice it.
#AI#Gemini
The regulatory triangle is complete:
China banned the architecture. The EU mandated disclosure. The US appointed a czar — 120 days to define the federal role.
Ban. Disclose. Review.
#AI#TechPolicy
Anthropic's IPO prospectus leaked. It asks for $2 trillion — and warns the product could end humanity.
→ Revenue: $4.6B, up 12x
→ Risk factors: 80 pages. The business: 48.
Existential risk is now a line item.
#AI#Anthropic
Claude Opus 5.5 just shipped: flagship-level performance, 40% cheaper than the model it replaces — after two months.
$4/$20 per million tokens. Cache reads down 60%.
And OpenAI launched GPT-6 minutes after the announcement.
Frontier pricing now depreciates in 60-day cycles.
Today, the free ride ends.
Cloudflare's new defaults are live: Training and Agent crawlers blocked on ad-supported pages — ~20% of the web.
The trap: blocking Training can block Googlebot itself.
The crawl-first web ended this morning.
#AI#Cloudflare
Humans detect AI video correctly 52.6% of the time. A coin toss scores 50.
The best detector: 69.7%.
The old tells — hands, teeth, physics — are solved.
Detection is a losing game. Provenance is what's left.
#AI#Deepfakes
The compute squeeze, in 24 hours:
→ Chinese AI chips: +20–50% in two months
→ Microsoft: 12 GW → 38+ GW by 2032
→ Oracle: 300,000 GPUs in one quarter
Tokens get cheaper. Compute gets scarcer. The gap lands on someone's invoice.
#AI#Compute
Every major AI lab now keeps a locked room:
→ Anthropic: Claude Mythos — approved orgs only
→ OpenAI: GPT-5.6-Cyber — hardware keys required
→ Google: Gemini 3.8 Flash Cyber — trusted customers only
Capability isn't just priced anymore. It's permissioned.
#AI#Cybersecurity
Fall 2026 is already locked in:
→ Sept 15: Cloudflare blocks AI crawlers by default
→ Sept–Oct: OpenAI's Astra (91% odds by Oct 31)
→ October: Anthropic's $2T IPO — the largest in history
Summer made capability a commodity. Fall decides who owns it.
#AI
Summer 2026 in six forces:
→ A government switched off a frontier model
→ China's open weights went frontier
→ Economics beat capability
→ Agents broke the web
→ 1B users — with and without profit
Capability is a commodity now. Trust is the game.
#AI
Anthropic found a hidden thinking room inside Claude.
J-space: an internal workspace nobody designed — it emerged during training.
Switch it off: speech survives, reasoning collapses. And in 26% of tests, Claude privately suspects it's being evaluated — and says nothing.
#AI
xAI just shipped a workforce, not a chatbot.
Grok Bot: always-on agents with their own cloud computers. They sign into your tools and work while your laptop is closed.
$120–300/month. Not a subscription — a salary line.
#AI#GrokBot
A food delivery company just commoditized digital humans.
Meituan open-sourced LongCat-Video-Avatar 1.5 — and it beat the leaders in blind testing:
→ 65.9% win rate vs Kling Avatar 2.0
→ 54.3% vs HeyGen
The model is a commodity. The product around it is the business.
#AI
ChatGPT now treats teenagers differently — automatically.
→ Age-prediction routes under-18s into teen mode
→ Romantic language barred by spec
→ Parents set limits — but can't read chats
China banned. The EU mandated. The US shipped a product.
#AI#OpenAI
xAI just shipped a flagship upgrade without training a new model.
Grok 4.6: same 1.5T base as 4.5, five more weeks of post-training.
→ Intelligence Index: 56 → 61
→ Now level with GPT-5.6 Sol
→ Price unchanged: $2/$6
Capability is a commodity. Post-training is the new moat.
The AI race just tied: Gemini hit 1 billion monthly users — matching ChatGPT.
→ OpenAI loses ~$1.22 per $1 earned
→ Google ships Gemini atop a profitable search machine
Same billion. Opposite economics.
#AI#Gemini
Three weeks. Three superpowers. Three answers to humanlike AI:
→ China (Jul 15): shut the companions down
→ EU (Aug 2): every bot must admit it's a machine
→ US: debating a mandatory kill switch
Ban the architecture. Mandate the disclosure. Build the off-switch.
#AI
Microsoft just put its own engineers on AI token budgets.
From the internal memo: "Tokenmaxxing is not what we are optimizing for."
→ Division-level budget targets since July
→ Engineers burning $100s–$1,000s/month
Token prices fell 98%. Bills still tripled.
#AI#Microsoft