You can’t even blink with these AI models. Every other day, you hear about “life-changing” new features. Most of it is exaggerated, but it’s still impressive. OpenAI and ChatGPT have the limelight today.
OpenAI has unveiled GPT-4.5, its most advanced language model to date. They say it has improved emotional intelligence and conversational depth.
CEO Sam Altman says GPT-4.5 is “the first model that feels like talking to a thoughtful person.” Being the salesman that he is, he says he was pleasantly surprised when he received genuinely good advice from the AI.
Key Features of GPT-4.5
Enhanced Emotional Intelligence
GPT-4.5 is trained to better understand and respond to social cues. This should make the AI’s interactions feel more natural and empathetic. It allows the model to provide nuanced support, such as offering comfort during challenging times.
I don’t know anyone looking for this kind of support, but it’s there now.
Here, OpenAI demonstrated how the new model handles a request to write an angry text to a friend. GPT-4.5 manages the situation better, as you can see.
Although, one could argue it performed worse because it strayed further from the request and tried to play the role of a responsible aide, stopping you from sending a text you might regret.
Listen, if you vent to your AI, then more power to you, I guess.
Reduced Hallucinations
GPT-4.5 has a lower tendency to generate inaccurate or fabricated information across various topics. Now, this is huge.
The model generates fewer inaccuracies compared to GPT-4. It has a lower hallucination rate of 37.1%, compared to GPT-4o’s 61.8%.
The AI hallucination problem persists, and any update that reduces it will always be welcome.
Improved Performance
As you would expect, with an expanded dataset and improved training techniques, GPT-4.5 offers deeper contextual understanding across diverse topics.
GPT-4.5 excels in tasks like writing, programming, and problem-solving, making it a versatile tool for both personal and professional applications.
Not Free for Now
However, these advancements come with increased computational demands. Altman says GPT-4.5 is a “giant, expensive model” and adds that GPU shortages have limited its immediate availability.
Currently, GPT-4.5 is accessible to ChatGPT Pro subscribers and developers via the API, with plans to expand to other subscription tiers as more GPUs become available.
Comments
2 responses
Hoping to find a way to obviate or at least mitigate that burden, Harari also initiated one other
particularly ambitious recruitment program. “I heard that the KGB used to identify orphans, who wouldn’t
have any restrictive commitments, at the age of thirteen or fourteen, took them under its wing, and trained
them under the best possible conditions to be operatives functioning under cover. I thought it was a good
idea.” He got Caesarea’s psychologist to find a fourteen-year-old Israeli boy without parents, and Caesarea
took him under its wing, without his knowing who was now taking care of him. A psychologist and two
tutors kept constant tabs on him, and he was given an excellent education, enrichment classes in art and
culture, and time for sports and leisure. “We told him we wanted him to serve the nation when he grew up,”
says Harari.
All of the grooming and tuition was a success. The boy became a soldier and then a talented young
officer, highly cultured and capable of functioning under cover as a foreigner. But the project as a whole
failed. “It became clear to me that it may work in a totalitarian country like Russia, but not in Israel. The
Israeli Jew doesn’t have that kind of perseverance, and very soon he wanted a girlfriend, a civilian career,
and a good salary. He would likely have aspirations in very different directions than what the intelligence
world could offer. We had no alternative but to just let him go his own way.”
Other recruitment policies proved more successful. For practical operational reasons, the Mossad
actually became an early pioneer in gender equality. “It is a huge advantage having a woman on the team
during an operation,” says Ethan. “A team made up of members of both sexes…always provides for a
better cover story and reduces suspicions.”
If recruits manage to make it past the initial screening stage, they then begin the “operatives course.”
Most Caesarea cadets never visit Mossad headquarters during their training and have no contact with other
trainees. They are exposed to as little information as possible so that they will have little to reveal if they are
captured and tortured. During training, their home base is one of several apartments in Tel Aviv, so that
they won’t be exposed to regular Mossad employees who come to work at headquarters every day.
Cadets undergo training in a host of spycraft fields—encrypted Morse code transmission (until
technological developments rendered that mode of communication obsolete), surveillance, losing tails,
weaponry, and hand-to-hand combat. They also study the geography, politics, and history of the Arab
countries.
Skills are honed in a series of practice missions, almost entirely on Israeli soil, and are usually mundane:
slipping a listening device into a telephone in a bank lobby, recovering innocuous documents, breaking into
homes or businesses just to prove a recruit has the skill.
The creation of an elaborate cover story goes well beyond a fake name. The recruit is expected to learn
an entirely new, fictitious biography: where he was born and raised, who his parents were, the social,
cultural, and economic background in which he was raised, his hobbies, and so forth. A passport issued by a
friendly base country—or forged by the Mossad—will allow the warrior to travel freely, even to countries
that don’t admit Israelis. The counterfeit profession, meanwhile, typically is one that requires frequent
international travel and lots of time spent working alone, without partners, offices, or regular hours. A
newspaper reporter or photographer, for example, or a screenwriter doing research for a movie script both
work well, because neither requires much explanation.
Great breakdown of GPT-4.5’s improvements! The reduced hallucination rate is definitely a game-changer. While some may prefer a more direct response, the AI’s responsible approach to emotionally charged requests seems like a step in the right direction. We recently explored how these advancements impact real-world applications—check out our https://modbibo.com/.