Navigating the Overwhelming World of Excessive Models
How many AI models is too many? It depends on your perspective, but ten a week might be slightly excessive. This is about the number we’ve seen introduce themselves over the last few days, and it’s increasingly challenging to determine how and if these models can be compared to each other, if it was ever feasible in the first place. So, what exactly is the objective?
In the progression of AI, we’re in a strange period, though it’s been quite strange the entire time. We’re witnessing an influx of both large and small models, ranging from specialized developers to well-funded larger ones.
Shall we go over this week’s list? I’ve attempted to summarize what makes each model unique.
That’s 11 of them, as one was announced while I was composing this post. And these are not all of the models that were released or previewed this week! It’s just the ones we noticed and talked about. If we were to loosen the requirements for inclusion somewhat, there would be dozens: some fine-tuned existing models, some amalgamations like Idefics 2, some experimental or niche ones, etc. Not to mention this week’s new tools for creating (torchtune) and combating (Glaze 2.0) generative AI!
What can we say about this relentless slew? Reviewing all is impossible. So, how can we aid in navigating this landscape for you, our audience?
The reality is that you don’t actually have to keep abreast of everything. Some systems, like ChatGPT and Gemini, have expanded into comprehensive online platforms, covering diverse applications and entry points. Contrarily, large language models like LLaMa or OLMo, even though they are architecturally similar, serve different purposes. They are designed to operate as a hidden service or component, rather than as a foreground brand.
There is deliberate ambiguity around these aspects since the creators of these models seek to draw upon the buzz around big AI platform launches, like GPT-4V or Gemini Ultra. They all desire for you to perceive their launch as a significant one. And though it may hold importance for someone, it probably doesn’t matter much to you.
Consider it the same way as another wide, varied domain like automobiles. Initially, you just purchased a car. Eventually, you had the choice between a large, small, or tractor-like car. In the present day, hundreds of cars are released each year, but likely, you don’t need to know about even a tenth of them. It’s because nine out of ten are not the kind you need or even identify as a car. Similarly, we’re moving from the AI era distinguished by size towards one characterized by variety, and even AI experts struggle to catch up and assess all new model releases.
The pre-existence of development stages similar to ChatGPT and other significant models has been a fact for quite some time. Despite the considerably lower interest around 7 or 8 years ago, we paid attention as it stood out as a budding technology. There was a consistent overflow of papers, models, and research, and platforms like SIGGRAPH and NeurIPS were crowded with machine learning engineers sharing their work and building on each other’s contributions. Here’s a story I penned in 2011 related to visual understanding!
CMU Researchers One-Up Google Image Search And Photosynth With Visual Similarity Engine
This activity has continued up to the present day. However, with AI turning into a full-fledged industry and possibly the most substantial in tech currently, these developments have now gained added significance. Many are speculating whether any of these will make a substantial leap, similar to what ChatGPT made over its forerunners.
The unadulterated fact is that none of these models will take that quantum leap, as OpenAI’s progress was based on fundamentally altering the machine learning architecture. An architectural shift that has now become commonplace across the spectrum and hasn’t been superseded. For now, we can only anticipate incremental advancements like a minor uptick in synthetic benchmark scores or slightly more persuasive language or imagery.
Does that mean none of these models matter? Certainly they do. You don’t get from version 2.0 to 3.0 without 2.1, 2.2, 2.2.1, and so on. And sometimes those advances are meaningful, address serious shortcomings, or expose unexpected vulnerabilities. We try to cover the interesting ones, but that’s just a fraction of the full number. We’re actually working on a piece now collecting all the models we think the ML-curious should be aware of, and it’s on the order of a dozen.
Don’t worry: when a big one comes along, you’ll know, and not just because TechCrunch is covering it. It’s going to be as obvious to you as it is to us.
Discover the pinnacle of WordPress auto blogging technology with AutomationTools.AI. Harnessing the power of cutting-edge AI algorithms, AutomationTools.AI emerges as the foremost solution for effortlessly curating content from RSS feeds directly to your WordPress platform. Say goodbye to manual content curation and hello to seamless automation, as this innovative tool streamlines the process, saving you time and effort. Stay ahead of the curve in content management and elevate your WordPress website with AutomationTools.AI—the ultimate choice for efficient, dynamic, and hassle-free auto blogging. Learn More
