State of Artificial Intelligence
@aisoni.bsky.social
Human perspective on AI
State AI Reports at www.stateofartificialintelligence.com
AI Insights www.aitechnologies.co/blog_menu.html
State AI Reports at www.stateofartificialintelligence.com
AI Insights www.aitechnologies.co/blog_menu.html
16/ I am sure even this year will be eventful.
#ai #artificialintelligence #technology #innovation #work #blockchain #law #quantum
#ai #artificialintelligence #technology #innovation #work #blockchain #law #quantum
February 4, 2026 at 3:10 PM
16/ I am sure even this year will be eventful.
#ai #artificialintelligence #technology #innovation #work #blockchain #law #quantum
#ai #artificialintelligence #technology #innovation #work #blockchain #law #quantum
15/ …for the bad (new GenAI extremely powerful malware is invented, serious cyber attacks are expected). Of course, this is not the only theme other significant factors are at play and we discuss them all with the full predictions in the report.
February 4, 2026 at 3:10 PM
15/ …for the bad (new GenAI extremely powerful malware is invented, serious cyber attacks are expected). Of course, this is not the only theme other significant factors are at play and we discuss them all with the full predictions in the report.
14/ As you may see, the few predictions above have a common theme: since the small LLMs now have more than capable performance, it is possible to unlock a series of applications for the good (local platform, countries with less resources can develop models, robotics benefits) or…
February 4, 2026 at 3:10 PM
14/ As you may see, the few predictions above have a common theme: since the small LLMs now have more than capable performance, it is possible to unlock a series of applications for the good (local platform, countries with less resources can develop models, robotics benefits) or…
13/ Chinese models will continue to dominate market share in low and middle income countries and the best on leaderboards for models <150B parameters.
There will be a serious drone/autonomous vehicle attack using AI which will raise awareness on vulnerabilities.
There will be a serious drone/autonomous vehicle attack using AI which will raise awareness on vulnerabilities.
February 4, 2026 at 3:10 PM
13/ Chinese models will continue to dominate market share in low and middle income countries and the best on leaderboards for models <150B parameters.
There will be a serious drone/autonomous vehicle attack using AI which will raise awareness on vulnerabilities.
There will be a serious drone/autonomous vehicle attack using AI which will raise awareness on vulnerabilities.
12/ As the next AI chip market growth will be due to the expected new robotics, there will be at least one player that announces a platform competitive in performance with Nvidia Jetson.
February 4, 2026 at 3:10 PM
12/ As the next AI chip market growth will be due to the expected new robotics, there will be at least one player that announces a platform competitive in performance with Nvidia Jetson.
11/ It will not be built by a major player (e.g. OpenAI etc.), get media attention and installable locally, not cloud based. At least 1M users (installation).
February 4, 2026 at 3:10 PM
11/ It will not be built by a major player (e.g. OpenAI etc.), get media attention and installable locally, not cloud based. At least 1M users (installation).
10/ Success rate of AI pilots in production (scaling) will be <15% for ‘complex/multi agents’ but above 30% for ‘narrow scope’ projects.
There will be an agentic platform low/no code (or framework) that gets traction on at least one industry or niche use case (not coding).
There will be an agentic platform low/no code (or framework) that gets traction on at least one industry or niche use case (not coding).
February 4, 2026 at 3:10 PM
10/ Success rate of AI pilots in production (scaling) will be <15% for ‘complex/multi agents’ but above 30% for ‘narrow scope’ projects.
There will be an agentic platform low/no code (or framework) that gets traction on at least one industry or niche use case (not coding).
There will be an agentic platform low/no code (or framework) that gets traction on at least one industry or niche use case (not coding).
9/ We will have a model of ~8B parameters or less with scores 20%+ on Humanity’s Last Exam Benchmark or similar (i.e. more than double current best intelligence performance of an 8B model).
February 4, 2026 at 3:10 PM
9/ We will have a model of ~8B parameters or less with scores 20%+ on Humanity’s Last Exam Benchmark or similar (i.e. more than double current best intelligence performance of an 8B model).
8/ As Nvidia’s dominance starts to erode (<85%), it will start a major lawsuit to limit CUDA portability into other chips.
February 4, 2026 at 3:10 PM
8/ As Nvidia’s dominance starts to erode (<85%), it will start a major lawsuit to limit CUDA portability into other chips.
7/ Anthropic Claude Code will still be the best coding LLM on major benchmarks and with more traction among coders.
On the top 10 model leaderboards for video or images there will be companies from at least 7 different countries during the year.
On the top 10 model leaderboards for video or images there will be companies from at least 7 different countries during the year.
February 4, 2026 at 3:10 PM
7/ Anthropic Claude Code will still be the best coding LLM on major benchmarks and with more traction among coders.
On the top 10 model leaderboards for video or images there will be companies from at least 7 different countries during the year.
On the top 10 model leaderboards for video or images there will be companies from at least 7 different countries during the year.
6/ …than 10% better than the closest open weights model; at least 20 points better than the best 2025 model in one LLM overall benchmark leaderboard released in 2025 (GDPval, Humanity’s Last Exam etc.); a new algorithm/approach to overcome ‘attention collapse’.
February 4, 2026 at 3:10 PM
6/ …than 10% better than the closest open weights model; at least 20 points better than the best 2025 model in one LLM overall benchmark leaderboard released in 2025 (GDPval, Humanity’s Last Exam etc.); a new algorithm/approach to overcome ‘attention collapse’.
5/ Best overall (i.e. best performance) ‘family of models’ will still be from a BigTech or Lab (>$1B) with: a hybrid architecture (not only transformer based); diffusion-based for ‘small/fast’ model size; proprietary, at least at the thinking level (best of the family); no more…
February 4, 2026 at 3:10 PM
5/ Best overall (i.e. best performance) ‘family of models’ will still be from a BigTech or Lab (>$1B) with: a hybrid architecture (not only transformer based); diffusion-based for ‘small/fast’ model size; proprietary, at least at the thinking level (best of the family); no more…
4/ Without further ado, for those of you who want to jump straight to the predictions here they are (more on the report, but here are the most important):
February 4, 2026 at 3:10 PM
4/ Without further ado, for those of you who want to jump straight to the predictions here they are (more on the report, but here are the most important):
3/ These range from semiconductor wars and agentic workflows to the frontiers of space data centers. And everything in between (energy potential issues, robotics, science breakthrough etc.).
If you’re interested, you can download it here.
If you’re interested, you can download it here.
February 4, 2026 at 3:10 PM
3/ These range from semiconductor wars and agentic workflows to the frontiers of space data centers. And everything in between (energy potential issues, robotics, science breakthrough etc.).
If you’re interested, you can download it here.
If you’re interested, you can download it here.
2/ At AI Technologies, we have just published our State of AI 2026 Report.
Reviewing the past year, it was insanely rich in events and rapid AI progress, which translated into an even longer report (it is getting longer year after year).
Reviewing the past year, it was insanely rich in events and rapid AI progress, which translated into an even longer report (it is getting longer year after year).
February 4, 2026 at 3:10 PM
2/ At AI Technologies, we have just published our State of AI 2026 Report.
Reviewing the past year, it was insanely rich in events and rapid AI progress, which translated into an even longer report (it is getting longer year after year).
Reviewing the past year, it was insanely rich in events and rapid AI progress, which translated into an even longer report (it is getting longer year after year).
11/ PS: this is true for many consultant professions: financial advisors, doctors lawyers etc.
#ai #artificialintelligence #future #business #technology #innovation
#ai #artificialintelligence #future #business #technology #innovation
January 28, 2026 at 11:50 AM
11/ PS: this is true for many consultant professions: financial advisors, doctors lawyers etc.
#ai #artificialintelligence #future #business #technology #innovation
#ai #artificialintelligence #future #business #technology #innovation
10/ That's why therapy or any other human interaction is irreplaceable.
Please use AI extensively even for self help, but trust what a professional sees on you beyond your own words.
I hope it helps.
Please use AI extensively even for self help, but trust what a professional sees on you beyond your own words.
I hope it helps.
January 28, 2026 at 11:50 AM
10/ That's why therapy or any other human interaction is irreplaceable.
Please use AI extensively even for self help, but trust what a professional sees on you beyond your own words.
I hope it helps.
Please use AI extensively even for self help, but trust what a professional sees on you beyond your own words.
I hope it helps.
9/ A professional who deserves that badge, will catch that, an AI no matter how excellent in replying (and it will be), it will not .
An AI as it stands now, cannot make any distinction between what a person says and how actually he feels.
An AI as it stands now, cannot make any distinction between what a person says and how actually he feels.
January 28, 2026 at 11:50 AM
9/ A professional who deserves that badge, will catch that, an AI no matter how excellent in replying (and it will be), it will not .
An AI as it stands now, cannot make any distinction between what a person says and how actually he feels.
An AI as it stands now, cannot make any distinction between what a person says and how actually he feels.
8/ From a professional point of view that's where the game starts: did the person lower his voice when said the word 'good'? Are the eyebrows raising?
Is he actually good when he 'said' he's good?
Is he actually good when he 'said' he's good?
January 28, 2026 at 11:50 AM
8/ From a professional point of view that's where the game starts: did the person lower his voice when said the word 'good'? Are the eyebrows raising?
Is he actually good when he 'said' he's good?
Is he actually good when he 'said' he's good?
7/ The AI gives you excellent advice throughout the conversation, then, at the end it will ask 'how do you feel now?'
The user will say 'I'm good now thanks'.
Here is the crucial moment:
From an AI perspective, that's it. The user 'said he's good', end of it.
The user will say 'I'm good now thanks'.
Here is the crucial moment:
From an AI perspective, that's it. The user 'said he's good', end of it.
January 28, 2026 at 11:50 AM
7/ The AI gives you excellent advice throughout the conversation, then, at the end it will ask 'how do you feel now?'
The user will say 'I'm good now thanks'.
Here is the crucial moment:
From an AI perspective, that's it. The user 'said he's good', end of it.
The user will say 'I'm good now thanks'.
Here is the crucial moment:
From an AI perspective, that's it. The user 'said he's good', end of it.
6/ Let me explain, the issue is not the 'quality of replies' but the 'quality of interaction'.
Quick example.
Let's say an user interacts with one of these self help AI apps about some trauma or past experience you are trying to resolve.
Quick example.
Let's say an user interacts with one of these self help AI apps about some trauma or past experience you are trying to resolve.
January 28, 2026 at 11:50 AM
6/ Let me explain, the issue is not the 'quality of replies' but the 'quality of interaction'.
Quick example.
Let's say an user interacts with one of these self help AI apps about some trauma or past experience you are trying to resolve.
Quick example.
Let's say an user interacts with one of these self help AI apps about some trauma or past experience you are trying to resolve.
5/ So here is the actual question: can these apps replace a psychologist, doctor or any other coach?
Well as an AI expert my answer is a... Resounding no.
The attentive reader may catch my line above that 'AI advice may be on par with professionals ', am I contradicting myself?
Well as an AI expert my answer is a... Resounding no.
The attentive reader may catch my line above that 'AI advice may be on par with professionals ', am I contradicting myself?
January 28, 2026 at 11:50 AM
5/ So here is the actual question: can these apps replace a psychologist, doctor or any other coach?
Well as an AI expert my answer is a... Resounding no.
The attentive reader may catch my line above that 'AI advice may be on par with professionals ', am I contradicting myself?
Well as an AI expert my answer is a... Resounding no.
The attentive reader may catch my line above that 'AI advice may be on par with professionals ', am I contradicting myself?
4/ They definitely can give the person immediate support with the undeniable benefit of not feeling judged.
And these apps have AI systems constantly improving that will give good advice to users possibly on par with professionals.
And these apps have AI systems constantly improving that will give good advice to users possibly on par with professionals.
January 28, 2026 at 11:50 AM
4/ They definitely can give the person immediate support with the undeniable benefit of not feeling judged.
And these apps have AI systems constantly improving that will give good advice to users possibly on par with professionals.
And these apps have AI systems constantly improving that will give good advice to users possibly on par with professionals.
3/ The company just released its first AI chatbot tool and got 50,000 beta users in 18 months.
Talking to someone about your issues, being listened to is definitely therapeutic.
Talking to someone about your issues, being listened to is definitely therapeutic.
January 28, 2026 at 11:50 AM
3/ The company just released its first AI chatbot tool and got 50,000 beta users in 18 months.
Talking to someone about your issues, being listened to is definitely therapeutic.
Talking to someone about your issues, being listened to is definitely therapeutic.