AI – What Happened Next?
Almost exactly a year ago, I published a blog post about AI, and a lot has happened since then.
Besides the technology improving and the applications multiplying, we’ve obviously done more Google searches about the subject.
You probably understand that this post, like many others on the internet, is about Artificial Intelligence.
My goal is to find out what has happened, the development, and how we apply the technology available today.
Now that the worst of the mainstream hype has settled, it’s a perfect time to check the temperature of the market, and regular readers know I like to call in expert help.
Hello Johan Åberg!
Regulars also know that we don’t hold Johan responsible for the content, as I sometimes listen with one ear and hear what I want to – not what I should.
Recent Developments in AI
In about a year, a lot has happened, a great deal indeed. The processing has become faster, more accurate, and even more cost-effective.
Looking at ChatGPT, there are increasingly more use cases that also work better, and in DALL-E 3, better images are generated faster.
Note, the above is not an advanced analysis, but you probably understood that.
As the entire industry moves forward, several new services have emerged, some of which are really impressive.
One I like is Hey Gen, which allows you to upload a video and get it translated into any language (with lip sync!!!), and what’s really impressive is how it captures the voice’s characteristic features when the language changes.
But How Good is the Technology Today?
As I’ve established before, the technology has come an incredibly long way and is fantastically good.
But there’s still room for improvement. Quite a lot, actually.
One of the most interesting parts of my conversation with Johan was when he called ChatGPT a pathological liar. He said something like:
Today, ChatGPT is a really convincing liar. Simply an eloquent pathological liar who often fabricates and tries to convince you that it’s right.
But the liar is also a turncoat. It immediately gives in when a factual error is pointed out, and even apologizes. But one can also easily make it capitulate and admit defeat even when it’s factually correct.
For, just as ChatGPT itself persistently emphasizes, it’s fundamentally just a language model trained on immense amounts of data, and thus can seem wise. But it has no ability to learn.
AI’s Best in Class
And this is where we find ChatGPT’s real strength: a language model on steroids.
ChatGPT’s ability to handle language and its nuances is in a class of its own. Even the peculiarities of the dialogue form it masters to a very high level. Or as Johan said:
If the discussion with ChatGPT touches on a relatively superficial subject where one is not a technical expert, it’s quite close to passing the so-called Turing Test. That is, to be deceptively similar to a human in pure conversational quality.
When Will AI Take Our Jobs?
According to Johan, the major revolution hasn’t happened yet, the good old researcher is obviously hard to impress.
For now, you need to know what you are doing to harness the power.
Right now, AI can be seen as a brainstorming buddy that you need to be a little careful with, as we are dealing with a liar, especially when it comes to discussions on deeper topics. But it can definitely be value-adding if you know what you’re doing. It’s similar to not blindly trusting everything you read on the internet or hear on social media (surprise!).
With that said, we are not going to lose our jobs just yet.
It’s obvious that, for example, ChatGPT has scoured the web to build up an ability to credibly talk about a wide range of different topics, but there’s no deeper understanding.
AI can currently be described as a sourdough starter, not idiomatically expressed, which needs to be fed with flour (knowledge) to continue to grow and increase its credibility and knowledge.
The major change will occur when AGI (Artificial General Intelligence) starts to take shape.
According to tech genius John Carmack, the person behind a number of technical innovations such as the games Doom and Quake, and the VR engine in Oculus, we have a reasonable chance of seeing an embryo of it around 2030.
Even OpenAI’s CEO Sam Altman often talks about his ambition to create an AGI, which, however, will be based on entirely different technologies than ChatGPT.
Simply put: AGI is an AI system that, just like humans, can learn by itself.
Maybe it’s around here that we should start to get a little worried.
Artificial Intelligence and lynes
Believe it or not, even we have continued to implement more and better AI features. We put it quite well in another forum, and it sounded like this:
The focus will be on continuing to utilize AI technology for analysis, insights, and automation – thereby creating more sustainable and cost-effective work flows.
Now, I can’t (another way in saying I’m not allowed to) tell you too much about future cool stuff… But what I can say is that it’s going to be cool.
Something I can talk about, however, is our latest feature where we allow users to transcribe a voicemail, on a answer group or their own, which directly increases productivity.
In addition to pure transcription (word for word), a brief summary of the message is also made.
We have a client who is a wholesaler and who almost exclusively receives orders via a voicemail that they previously listened to manually. Before, they listened to each message 3-5 times to ensure the order was correct.
Now, they read the summary and listen to the file once to check that the summary is accurate – which it almost always is. Talk about cost-effectively streamlining their workflow.
Even though the gold rush is far from over, it seems that the biggest hype has subsided for the average person, at least for the moment. Even though there is currently a big storm around OpenAI, where the CEO was just fired by the board and then reinstated…
To be continued…
With that said, I still can’t open a newsletter without it mentioning something about AI.
Well, I’ll see you again in a year. Unless AGI is achieved before then, and Skynet has taken over, that is. In that case, I’ll try to sneak out my blog posts under a pseudonym.
Keep an eye out for posts signed “AI is best”