Welcome to 2025 – Ai News

The last year in Ai has been a real roller coaster ride, with twists and turns, resulting in massive changes in Ai development with its progress accelerating at an exponential rate.

Possibly the most important image today is the one that shows the evolution of Human intelligence compared with that of Artificial intelligence.

This chart shows the rise of Human intelligence over centuries compared with the rise of Artificial intelligence from some time midway in the last century.

At a point in time the chart shows ‘The Singularity’, the crossover when Ai exceeds that of Human intelligence. It means that at that point, Ai has learned everything that it can from Humans and has matched our best intelligence (AGi – Artificial General Intelligence)). The next move is to go beyond the Singularity to a time when computers have progressed so much that they are able to self-learn, innovate, and create technology that’s entirely new, and foreign, to humans (ASi – Artificial Super Intelligence).

ASi is when systems start to massively exceed human abilities to even fully understand their working. Many commentators believe that the singularity point is well in the future, perhaps 2045 or even maybe not until the next century.

They might be wrong.

Just a short while ago (4^th Jan 2025) Sam Altman, the CEO and one of the original founders of OpenAi, started the year with a tweet that simply read:

i always wanted to write a six-word story. here it is:
‘near the singularity; unclear which side.’

To give some context, in December 2024 OpenAi’s model o3 produced some staggering results in a series of benchmark tests designed to test Artificial General intelligence (AGi). And it came out just when some were predicting an Ai winter – where Ai progress hits a wall and stops.

Clearly that hasn’t happened, and since then several senior people who are directly involved in developing Artificial intelligence are discussing the possibility of ASi being quite close.

So, the big question for 2025 is not, ‘What are the latest gismos for movie making on my PC’, but rather ‘Where are we on the curve?’

We shouldn’t expect a big sign to appear when we cross the Singularity, but as Sam so eloquently writes – it’s hard to know which side of it we are currently on. We do seem to be in the vicinity.

We are all living through a pivotal time in human history. Savour the moment, and watch out for some big developments in 2025!

No sooner than the proverbial ink had dried on those electronic words, that we received an onslaught of Ai system releases in 2025!

Chinese company Deepseek launched a range of market disruptors:
January 10th Deepseek V3 released
January 20th Deepseek R1 family released
January 28th Deepseek Janus Pro (multi modal model)

All of these are powerful, fully open source, Ai systems competing with the best of the rest. They were launched with the claim that their training was a fraction of the cost of the main competition. This alarmed investors into thinking that maybe Nvidia and other tech companies didn’t need their massive investments, causing stock prices to drop on 27th January.

During this time other Chinese companies such as Baidu, Btye Dance, Tencent, and Alibaba (Qwen) all released their models showing to a surprised Western World that China is a very close rival to the US in Ai.

US companies weren’t slow to respond. On 30th January, the Allen Institute for Ai (Ai2) released its Tulu3 405B reasoning model. Based in Seatle this little known non-profit institute’s model is also fully open source and matched Deepseek as a comparable reasoning model.
On 31st January, OpenAi’s more powerful and recently announced high end reasoning model o3-mini was made public.

Then on the 2nd February, Open Ai released Deep Research, an Ai agent capable of multi-step research on the internet for complex tasks. Just to cause a bit of confusion, the name is similar to that of Deepseek and is the same as the Google’s Gemini Deep Research offering from last year. OpenAi’s model is powered by their most powerful model – o3 (not to be confused with their o3-mini released a few days ago) and places it firmly at the top of the performance charts (for today!).

As per the OpenAi Deep Research release notes:

Many believe that humans last great invention will be AGi, because after that it will be the computers that will do all the inventing. That’s why Ai safety is so critically important, and why we are entering uncharted territory.

’Wow’! What a start to 2025 in Ai!

Now take a breath while you consider a few things:

When something is free to you, you are probably the product. Consider what data is being collected? Where is your data is going? How is it being used? What are your privacy rights? US products connect to US servers, Chinese products connect to Chinese servers.
Deepseek is connected with a investment company that created and used Ai stock trading systems for the markets. It should have been obvious to anyone with stock market experience the impact Deepseek products would have on stock prices. It’s reasonable to wonder if any of their investors had the good sense to short the technology stocks? Out of interest, Reuters reported that $6.6Bn was made by Nvidia short investors.
Assuming it’s true that much less compute was required compared to the US models, the lower costs are more likely to drive demand up, rather than down, for products such as Nvidia’s. Lower costs have always resulted in greater demand.
USA restricting sales of tech (such as Nvidia H100 GPU’s) to overseas countries often results in further proof that ‘Necessity is the mother of invention’.
Models still tend to hallucinate with their answers, but bear in mind that these small hallucinations are the last thin line of defence for Ai taking much of the white collar work.
Not everyone tells the truth !