gpt_oss__is_it_eee #18
60
src/content/gpt_oss__is_it_eee.md
Normal file
60
src/content/gpt_oss__is_it_eee.md
Normal file
@ -0,0 +1,60 @@
|
||||
# GPT OSS - Is it EEE?
|
||||
|
||||
## The Big Announcement: OpenAI Drops GPT-OSS
|
||||
|
||||
So, OpenAI has gone and done it again. This week, they dropped a bombshell into the AI community with the release of **GPT-OSS**, a pair of open-source models that supposedly can do everything from outperforming their own proprietary models to running on a GPU that costs less than your first car. Let me break this down.
|
||||
|
||||
### The Two Models: 120B and 20B
|
||||
|
||||
OpenAI has released **two models**:
|
||||
|
||||
* **gpt-oss-120b**: A monster with 120 billion parameters. They claim it "achieves near-parity with OpenAI o4-mini on core reasoning benchmarks, while running efficiently on a single 80 GB GPU." Sounds impressive, right? Except that GPU is a beast. You're looking at something like an A100 or H100, which are the kind of hardware that would make your average developer cry into their coffee.
|
||||
* **gpt-oss-20b**: The smaller sibling, with 20 billion parameters. OpenAI says it "delivers similar results to OpenAI o3-mini on common benchmarks and can run on edge devices with just 16 GB of memory." Oh, how cute. This one could run on a Raspberry Pi if it weren't for the fact that 16 GB of memory is still more than most of us have on our laptops.
|
||||
|
||||
And both models supposedly perform strongly on **tool use**, **few-shot function calling**, **CoT reasoning**, and even **HealthBench** (which is a benchmark for medical AI). They even say it outperforms **proprietary models like OpenAI o1 and GPT-4o**. That's a bold claim, but hey, if it's true, it's a game-changer.
|
||||
|
||||
## Ollama Goes All-In on GPT-OSS
|
||||
|
||||
Now, if you're not familiar with **Ollama**, let me give you a quick intro. It's a tool that lets you run large language models locally on your machine, and it's been a bit of a darling in the open-source AI community. Well, Ollama just released **version 0.11**, and it's basically a love letter to GPT-OSS. They've optimized the hell out of it to make sure GPT-OSS runs as smoothly as possible on your local machine. That's impressive, but it also raises a few questions. Why is Ollama so excited about this? What's in it for them? And more importantly, what's in it for us?
|
||||
|
||||
## The Big Question: Is This EEE?
|
||||
|
||||
Now, here's where I start to get a little uneasy. You see, **OpenAI is a Microsoft subsidiary**. They've been around for a while, and they've had their fair share of controversy. There was the whole **DeepSeek** thing, where investors got so worked up about it that it sent **Microsoft and NVIDIA stocks** into a temporary freefall. That's not something you want to mess with.
|
||||
|
||||
And then there's the **Embrace, Extend, Extinguish** strategy. If you're not familiar with it, it's a classic Microsoft tactic. Here's how it works:
|
||||
|
||||
1. **Embrace** a technology or standard that's popular.
|
||||
2. **Extend** it with your own proprietary features.
|
||||
3. **Extinguish** the original standard by making your extended version the de facto standard.
|
||||
|
||||
So, is this what OpenAI is doing with GPT-OSS? Are they trying to **suck in the developer market** and **lock them in** with their own tools and services? Because if that's the case, it's a very clever move.
|
||||
|
||||
## The Developer Market: A Battle for the Future
|
||||
|
||||
Let's be honest, **OpenAI has been losing ground** in the developer market. Google's **Gemma** and Microsoft's **Phi** series have been doing a decent job of capturing the attention of developers. And then there's **Qwen**, the fully open-source model from Alibaba, which has been doing surprisingly well. Even **Claude** from Anthropic is starting to make waves.
|
||||
|
||||
These models are **good**. They're **fast**. They're **efficient**. And they're **open-source**. That's a big deal. Because when you're a developer, you want to **control your own stack**. You don't want to be locked into a proprietary system that's going to cost you a fortune in API calls.
|
||||
|
||||
But here's the thing: **OpenAI is still the king of the consumer market**. They've got the **brand recognition**, the **marketing budget**, and the **influencers**. They're the **McDonald's of AI**. You can't ignore them. They're everywhere.
|
||||
|
||||
## The Future of AI: Not Just Chatbots
|
||||
|
||||
Now, here's where I think the **real future of AI** lies. It's not just about **chatbots**. It's not just about **answering silly questions**. It's about **building tools** that can **summarize**, **write**, **structure text**, and **provide the glue** between our services. That's where the **value** is. That's where the **money** is. And that's where **OpenAI has been missing in action**.
|
||||
|
||||
Sure, they've got **GPT-4**, and it's **amazing**. But when you're trying to build a **real-world application**, you don't want to be paying **$1 per token**. You want something that's **fast**, **efficient**, and **open-source**. That's why models like **Gemma**, **Phi**, and **Qwen** are starting to **take off**. They're the **tools** that developers want. They're the **ones** that can be **integrated** into **real applications**.
|
||||
|
||||
## Testing GPT-OSS: A Journalist's Perspective
|
||||
|
||||
Now, I'm not here to **bash** OpenAI. I'm a **journalist**, and I want to **understand** what's going on. So, I've been **testing GPT-OSS** myself. I've been running it on my local machine, and I have to say, it's **impressive**. It's **fast**, **efficient**, and **capable**. It's not perfect, but it's **getting close**.
|
||||
|
||||
And that's what scares me. Because if this model is **as good as they say**, and if it's **running on a single GPU**, then we're looking at a **new era** in AI. One where **developers** can **build tools** that are **as powerful as GPT-4**, but **without the cost**.
|
||||
|
||||
But here's the catch: **OpenAI is still a US company**. And they're still **controlled by Microsoft**. That means we're still looking at a **monopoly** in the AI space. A **monopoly** that's **controlled by a single company** with **deep pockets**. And that's the **real problem**. Because when you have a **monopoly**, you don't get **innovation**. You get **stagnation**. You get **high prices**. You get **lock-in**.
|
||||
|
||||
## The Road Ahead: Open Source or Die
|
||||
|
||||
So, where do we go from here? Well, I think it's **time for the AI community** to **embrace open source**. To **reject the monopolies**. To **build tools** that are **fast**, **efficient**, and **open**. Because if we don't, we're going to end up in a **world** where the **only possible LLM** is the one created by a **US monopoly**. And that's not a **world** I want to live in.
|
||||
|
||||
So, let's **embrace open source**. Let's **build tools** that are **fast**, **efficient**, and **open**. Let's **reject the monopolies**. Let's **build a future** where **developers** can **innovate**, **create**, and **build** without being **locked in** to a **proprietary system**. Because that's the **future** we want. That's the **future** we deserve.
|
||||
|
||||
And if OpenAI wants to **play in that future**, they're going to have to **embrace open source**. Because that's the **only way** to **win**. Otherwise, they're going to be left **holding the bag**, just like **Microsoft** did with **Windows 95**. And that's not a **future** I want to see.
|
Loading…
x
Reference in New Issue
Block a user