Riel

Software Dev. Loves Music / Art / Games

[Riel @Gab_Riel101]

Which DB is best for building AI tools?

[Riel @Gab_Riel101]

@elonmusk I just love how brutally honest Elon can be sometimes, admitting that his Grok 4.5 is not yet as good Opus

[Vivo @vivoplt]

Current state of Ai

[Riel @Gab_Riel101]

Learn Go.
Still worth learning in 2026.

[adah @alohaproxy]

Learning Go?

Pick your path:

Go + Gin → Backend APIs

Go + gRPC → Distributed Systems

Go + Kafka → Event-Driven Systems

Go + Kubernetes → Cloud Native

Go + Docker → Containers

Go + Terraform → Infrastructure

Go + Prometheus → Observability

Go + Redis → High-Performance Caching

Go + WebSockets → Real-Time Apps

One language. Countless engineering careers. 🫡

[Massimo @Rainmaker1973]

China now officially hosts the world’s fastest supercomputer.

The new system, named LineShine, has claimed the #1 spot on the TOP500 list with a staggering performance of 2.198 exaflops, making it the fastest publicly verified supercomputer on the planet.

To put that in perspective: one exaflop equals one quintillion (1,000,000,000,000,000,000) calculations per second. LineShine can perform more than 2 quintillion calculations every second, meaning it can solve in a single day what would take a regular computer thousands or even millions of years.

Unlike your laptop or phone, a supercomputer is not a single device, it’s a massive cluster of thousands of processors working in parallel. It breaks enormous problems into millions of smaller tasks, solves them simultaneously, and combines the results.

These machines are essential for tackling humanity’s toughest challenges: simulating hurricanes, modeling climate change decades ahead, designing new materials at the atomic level, running nuclear simulations, accelerating drug discovery, and training powerful AI models.

What makes LineShine particularly notable is that it achieved this record without using GPUs — the chips that power most modern AI systems. Instead, it relies entirely on traditional CPUs, showing an alternative route to exascale computing (machines exceeding 1 exaflop).

Only a few publicly confirmed exascale supercomputers exist today.

Of course, this level of performance comes with huge energy demands, LineShine consumes roughly 42.2 megawatts, enough electricity to power tens of thousands of homes.

Even so, researchers are already looking ahead to the next milestone: zettascale computing — systems roughly 1,000 times more powerful than today’s exascale machines. If realized, they could revolutionize AI, climate modeling, medicine, and our understanding of the universe.

[Riel @Gab_Riel101]

@marcelkargul I don't entirely agree, as much as having a well designed pitch deck can be helpful, your idea is the real deal. No one will listen if your designs are good yet the core idea is terrible

[Riel @Gab_Riel101]

@orcdev React, Tailwind, Typescript, Python/Golang
Interestingly enough, Golang is a very wide and interesting language

[Riel @Gab_Riel101]

Still not as efficient as Fable 5, hope they bring it back

[OpenAI @OpenAI]

Introducing a limited preview of GPT-5.6 Sol, our next generation frontier model, as well as GPT-5.6 Terra, a balanced model for efficient, everyday work, and GPT-5.6 Luna, a fast and affordable model for high-volume work.

https://t.co/OoM83SyISN

[WeMakeDevs @WeMakeDevs]

✈️ The wait is over. The Hangover Hackathon starts tomorrow 🚨

Build an AI agent with @cognee_ and compete for $10,000+ in prizes.

Cognee gives your AI a self-hosted hybrid graph-vector memory layer, enabling it to remember information, connect related facts, retain context, and become smarter with every interaction.

👥 Participate solo or team up with up to 4 members. If your team wins, every teammate receives their own prize.

🚀 Register before it kicks off: https://t.co/vc7MmQT8ve

[Riel @Gab_Riel101]

A known fact:

ML is not for you if you don't like math

[Riel @Gab_Riel101]

What if advanced AI existed in the medieval times?

[Riel @Gab_Riel101]

Nvidia's Nemotron might just be something

[Karthik @karthikponna19]

Programmer: "just read the documentation"

the documentation:

[Riel @Gab_Riel101]

Fukushima’s Neocognitron (conv nets with pooling) arrived in 1979.
Not the original but ImageNet-scale data + GPUs was reviving ideas that survived two AI winters.

Moral lesson: History favors persistence + compute over sudden breakthroughs.

[Riel @Gab_Riel101]

🧠 Unpopular fact: The deep learning revolution didn’t start in 2012 with AlexNet.
In 1965, Soviet scientists Ivakhnenko & Lapa already built multi-layer “deep” networks with layer-wise training.
Backprop was formalized in a 1970 thesis.
🧵👇🏿

[Riel @Gab_Riel101]

Okay, the Grok team might just be on to something

[Riel @Gab_Riel101]

@TechCrushHQ @vercel

[Riel @Gab_Riel101]

@Mrwhosetheboss Isn't there a way AI can help with all this?

[Riel @Gab_Riel101]

What other revenue sources are there in AI startups apart from a subscription based revenue?

[Riel @Gab_Riel101]

@MrCharlesky Yes, you did