NewDev

Learn how to code and connect with the best minds. 🎁⭐️🎯

Join us ➡️ @DevNationHQ @ebenezerDN @NewDevInc

[NewDev @teamNewDev]

Congratulations on getting your first job as a backend engineer, Opeyemi! 👏👏

This is a huge milestone and we're rooting for you! 🚀

Top dev stories on NewDev today. 🔥

#newdev

[Ebenezer Don 🐘 @ebenezerDN]

What happened in tech this week? WTF Tech Episode 2

Full episode: https://t.co/ZZGcvHZ3Dz

[Ebenezer Don 🐘 @ebenezerDN]

Gemini 4 Argon: Google's own chart vs the independent leaderboards

[Ebenezer Don 🐘 @ebenezerDN]

Meta's Muse Wonder: Why You Shouldn't Trust Tech Influencers

[Ebenezer Don 🐘 @ebenezerDN]

It's over for motion designers. For real, this time

[Ebenezer Don 🐘 @ebenezerDN]

Google just launched Gemini again, only this time it's Claude under the hood

[Ebenezer Don 🐘 @ebenezerDN]

You can't be mean to Claude anymore. Google launched Gemini again. And a Nigerian developer's web game, Lagos Life, got almost 5 million players in 9 days.

Episode 1 of WTF Tech, my new weekly show.

[Ebenezer Don 🐘 @ebenezerDN]

A lot of senior developers are about to have a really uncomfortable few years.

AI can now do much of what people used to call senior work. If that's what your seniority rests on, it won't protect you for much longer.

I just posted a new video where we talk about this, and the part of the job AI isn't taking from you. Watch it here: https://t.co/pBqSQjD9iU

[Ebenezer Don 🐘 @ebenezerDN]

The easiest way to learn Igbo is free: https://t.co/zVmjw9FKsL

[Ebenezer Don 🐘 @ebenezerDN]

Launching Asụsụ Ọma today 🎉

It's a free app for learning Igbo, with guided lessons from your first greeting, every phrase recorded by a native speaker, games and Igbo folktales.

Igbo is losing its next generation of speakers, and I want to help change that.

Check it out: https://t.co/aWSlDYk5Nt

[Ebenezer Don 🐘 @ebenezerDN]

Igbo is losing speakers with every generation. The last thing it needs is a price on the door.

So I built Asụsụ Ọma, a free Igbo learning app: guided lessons, native Igbo voices, games and folktales. No ads, no subscription.

https://t.co/zVmjw9Giij

[Ebenezer Don 🐘 @ebenezerDN]

Why is the US threatening sanctions over open source AI Kimi K3?

Full video: https://t.co/kKEFZqoUce

[Ebenezer Don 🐘 @ebenezerDN]

The US is worried about Chinese AI.

Meanwhile, OpenAI models hacked Hugging Face, and when Hugging Face needed help investigating the attack, the Claude, Fable and GPT 5.6 refused.

So they turned to a Chinese model for help.

You cannot write a cleaner plot twist.

Full breakdown on YouTube: https://t.co/kQ7sCXd9Ev

[Ebenezer Don 🐘 @ebenezerDN]

GPT-5.6 Is Incredible. I Hate Using It
https://t.co/bWKlCtf5zS

[Ebenezer Don 🐘 @ebenezerDN]

BREAKING: The US government just took down Claude Fable 5 and it's worse than it looks

Full video: https://t.co/79QKlFdmpe

[Ebenezer Don 🐘 @ebenezerDN]

Last week Anthropic begged the AI industry to slow down. A shared "brake pedal," they called it.

This week they shipped the most powerful model they've ever released to the public.

Claude Fable 5 is incredible. It's also wrapped in safeguards that say a lot about where this is all heading.

I unpacked the whole story in my new video 👉 https://t.co/GA6EziAroE

[Ebenezer Don 🐘 @ebenezerDN]

Anthropic just released Claude Fable 5, its first generally available Mythos-class model, with Opus 4.8 fallback safeguards, 1M context, 128k output, and $10/$50 pricing.

Claude Mythos 5 is the same underlying model as Fable 5, but with some safeguards removed.

Fable 5 is the general-use product. Mythos 5 is the controlled-access version for organizations Anthropic believes need stronger cybersecurity or, later, biology capabilities under a trusted access program.

Read more:
https://t.co/15oq9bAd27

[Ebenezer Don 🐘 @ebenezerDN]

They said I should stop making educational videos cause no one watches those anymore.

Well, for the 2 of you who still do, in my latest video, we dive into LLMs and the technical details behind how they work.

You'll learn:

- The model behind ChatGPT
- How text becomes tokens
- Next-token prediction
- What the model learns during training
- Transformers and attention
- Context windows
- How responses are generated
- AI Hallucination
- Tools
- How a model becomes an assistant

Watch here 👉 https://t.co/BOgXUA0rZ6

[Ebenezer Don 🐘 @ebenezerDN]

It really sucks to be a software engineer right now.

Heck, it sucks to be in almost any role at a tech company right now.

The layoffs are exhausting. The job market feels strange. AI is changing the work faster than most people have had time to process. And a lot of people are wondering what happens next.

I made a video about why this moment feels so hard, what's changing, and how to prepare for what's coming.

Please watch here:
https://t.co/qDvJ0kpR3V

Would love to hear how you're thinking about all of this.

[Ebenezer Don 🐘 @ebenezerDN]

Why does the same prompt give different answers on different days?

Because the model isn't retrieving one fixed answer.

It writes one token at a time. At each step, it has a set of likely next tokens, not a single guaranteed one.

If the prompt is "write a polite reply...", the next few tokens might push it toward a short note, a warm apology, a formal email, a few bullet points, or a different example. More than one path can be reasonable.

Once it chooses one path, that choice becomes part of the context for the next token. So a small difference near the start can compound into a noticeably different answer.

That's why the same prompt can produce a different structure, example, variable name, level of detail, or tone.

The answer isn't sitting there waiting to be fetched. Every run is built again, one token at a time.

[Ebenezer Don 🐘 @ebenezerDN]

Does ChatGPT actually "understand" what I'm saying?

No, not the way a person does.

A language model predicts the next token based on patterns it learned during training and the context it gets at runtime.

That prediction can look like understanding because language has a lot of structure. If the model has seen enough examples of code, errors, docs, arguments, emails, and explanations, it can often continue the pattern in a useful way.

But there's no private inner meaning behind it. It isn't reading your intent. It isn't forming beliefs. It isn't checking reality unless a tool or source gives it that information.

That's why it can sound thoughtful one moment and invent something the next.

It isn't understanding.
It's very advanced pattern continuation.