We’re on day 105 since Chat GPT arrived and things will never be the same. From initiating new spikes in plagiarism to writing an episode of South Park to sending millions into both panic & utopian fervor, there’s a whole list of reasons every investor is scrambling with y2K-era chaotic energy to get in on the AI craze. And it doesn’t seem like the hype is slowing any time soon.
On March 14th, OpenAI unveiled their new creation, GPT-4. Its improvements are, for a lack of better words, impressive. Here’s what you should know about the latest update:
IT’S EVEN SMARTER:
ChatGPT just underwent an intense 80s-themed training montage where it hit the books hard. For one, GPT-4 has absorbed the souls of Dr. Seuss, Shakespeare, and Steinbeck to upgrade its creativity.
While GPT-3.5 was undoubtedly witty, logical problems tended to send it into confusion. No longer. The child’s grown into a monster.
Oh, and if you thought Chat GPT was smart before, it’s raised its scores in all academic fields. Notably, where it once scored in the bottom 10% of those to take the Bar Exam, now it’s in the top 90%. Cue the intro to “Push It to the Limit.”
IT CAN ANALYZE IMAGES:
Another revolutionary addition to the GPT technology is its ability to interpret images along with text. Open AI stated, “GPT-4 is a large multimodal model (accepting image and text inputs, emitting text outputs) that, while less capable than humans in many real-world scenarios, exhibits human-level performance on various professional and academic benchmarks.”
Basically, the supercomputer now has eyes. Soon, we might all be able to ask GPT “how do I look?” and after five seconds of careful calculation, it will generate the perfect answer.
IT CAN GET DEEPER INTO CHARACTER:
Next is an improvement to what OpenAI deems “Steerability.” This is essentially the ability to set its tone, worldview, or limitations to customize the experience. Essentially, you can now assign Chat GPT an anchor of sorts (by describing those directions in the “system” message). It’s kind of like the tokens in the movie Inception that keep dream intruders in touch with what’s real, or like that episode of SpongeBob where he’s instructed to only focus on “Fine dining and breathing.”
And yet… IT’S STILL NOT PERFECT:
There are a few reasons GPT-4 is still flawed.
- “Alignment” the idea that AI’s intentions will align with humanity’s best interests is still yet to be fully understood. OpenAI stated, “Unaligned AI could pose substantial risks to humanity,” saying, “There is currently no known indefinitely scalable solution to the alignment problem. As AI progress continues, we expect to encounter a number of new alignment problems that we don’t observe yet in current systems.” Basically, as smart as GPT-4 is, it’s still not able to analyze its most existential flaw, potentially escaping the grasp of humanity.
- While significantly more accurate (40% more likely), GPT-4 can still be prone to “hallucinations” or giving false information or buggy code. For this reason, it’s not advised to use GPT technology for high-stakes situations without human intervention.
- GPT operates with limitations, however, it’s still possible to “jailbreak” these boundaries, such as violent and inappropriate remarks. It’s in these moments of jailbreaking that we can see GPT’s potential threat.
IT AIN’T FREE:
Sadly, AI is so in demand, they’ve opted to put GPT-4 behind a paywall. However, for only $20 a month, you get access to ChatGPT Plus, the world’s greatest, most up-to-date super assistant with lots of perks including the new update.
As for image analysis, this function isn’t even accessible to ChatGPT Plus users. There’s a waiting list, but if you’re lucky, conduct research, or perhaps run a large business, you can jump to the front.
Only 105 days out from launch, OpenAI is proving it has no plans to take its foot off the gas. Stay tuned for updates. There will be more.