This Is Why AI Hallucinates

In partnership with

Hey there,

This week, we saw the perfect example of a disconnect happening. 

OpenAI, Oracle, and SoftBank announced they're spending $500 billion on new AI data centers. 

But on the other side, 95% of enterprise AI implementations are failing. Companies with over 10,000 employees are losing up to $9 million annually to "AI workslop", polished-looking AI output that actually makes work harder, not easier.

So half a trillion dollars is being thrown at infrastructure, but most companies aren’t making any profit using AI.

It's like watching people talk about buying faster cars while most drivers are still crashing in parking lots.

So let's talk about what's actually happening with AI right now, the failures, the surprises, and the few things that are genuinely working.

The Free Models That Match GPT-5

China is once again proving that you don’t need big, expensive models to compete in this AI space.

Alibaba dropped Qwen3-Max, scoring 100% on AIME25 and HMMT25 benchmarks, matching GPT-5 level performance. But they're giving it away for free.

Then they released Qwen-3-Omni, the first truly omnimodal model that handles text, images, audio and video without converting between formats. It supports 119 languages, processes 30-minute audio files, and responds in just 211ms.

Their text-to-speech model? 37ms latency with 14 different voices in 10 languages.

China is building global adoption by making advanced AI free while western companies charge thousands for similar capabilities.

The result? Users worldwide are switching because these models work just as well and cost nothing.

Small AI Teams Beat Big Models

Andrew Ng recently made five predictions about AI's future.

His key insight: the future isn't about building bigger LLMs. It's about "agentic workflows". Smaller, cheaper models working together as a team.

Think of it like this: instead of hiring one genius to do everything, you hire a team of specialists who are really good at their specific jobs. JPMorgan already cut costs by 30% using this approach.

Ng's prediction that matters most: smaller AI models running on your own devices will be the next big shift. Not $500B data centres in Texas, but smart software on your laptop that doesn't need the internet to work.

This changes everything. If he's right, the companies spending billions on massive infrastructure are building the wrong thing entirely.

The Truth About AI Hallucinations

OpenAI admits “AI hallucinations are mathematically inevitable.” Which means no matter what they do, AI will always hallucinate stuff, even if just a little bit.

They're not bugs to be fixed – they're built into how these systems work.

This actually helps explain the failure rate. Companies that blindly rely on AI will suffer losses. The smart ones accept that AI makes mistakes and plan for it. 

They use AI for first drafts, not final decisions. They check important stuff and let AI handle the boring work.

The companies still waiting for flawless AI are the ones wasting millions on failed projects.

Creative Industries Are Getting Steamrolled

Music just got completely disrupted. Suno launched their updated version Suno 5, which can create songs that sound exactly like professional musicians. 

You give it a photo or a video clip, and it makes a song. Any genre. You can edit every note and add real-sounding vocals.

But it's not just music. Meta also launched an AI dating assistant in Facebook that lets you search for anything like "a Brooklyn tech bro who would go to EDM concerts with me" and actually finds matches. 

Their "Meet Cute" feature picks one person for you each week – no swiping.

Dating app stocks crashed immediately. Match Group (Tinder, Hinge) dropped 5%. Bumble fell 4%.

This shows the pattern happening everywhere. AI isn't just making existing things cheaper. It's completely changing how entire industries work. Music creation, dating, content – all getting flipped upside down.

The creative workers who are surviving aren't fighting AI. They're using it to do the boring stuff faster so they can focus on the human parts – emotion, connection, live performance.

What Does This Mean For You?

Here's what successful AI implementations actually look like:

• Stop chasing the biggest models. The companies winning with AI use specialized, smaller models that solve specific problems well. A $20/month tool that saves your team 10 hours a week beats a $10,000/month enterprise solution that nobody uses.

• Build teams of AI agents. Multiple focused AI tools working together often outperform single large models. Use one AI to research, another to write, and a third to fact-check. Each one costs less and works better than trying to make one AI do everything.

• Plan for mistakes. The companies succeeding with AI expect it to be wrong sometimes. They build workflows where humans verify important outputs and AI handles the grunt work. They don't trust AI with final decisions, but they let it handle first drafts.

• Think local, not cloud. The real breakthrough might be AI that runs on your devices instead of sending everything to distant servers. Faster, more private, and no monthly subscription fees.

The Real Question

The companies winning with AI aren't spending the most money on the fanciest models. They're solving real problems with the right-sized solutions.

$500 billion is being spent on infrastructure, but the real question isn't "How big can we go?" It's "What problems actually need solving, and what's the simplest way to solve them?"

Most AI failures happen because people try to use a Ferrari to deliver pizza. The tool is impressive, but it's the wrong tool for the job.

Are you building solutions for problems that will still exist in six months? Or are you chasing the infrastructure everyone else is building?

The AI race isn't won by the biggest spender. It's won by the smartest implementer.

What are you implementing?

-Aashish

P.S. If you're tired of AI hype and want to discuss what actually works, join our AI community. Real practitioners, real results, no BS.

Typing is a thing of the past

Typeless turns your raw, unfiltered voice into beautifully polished writing - in real time.

It works like magic, feels like cheating, and allows your thoughts to flow more freely than ever before.

With Typeless, you become more creative. More inspired. And more in-tune with your own ideas.

Your voice is your strength. Typeless turns it into a superpower.

Download for Mac today