A dispute over overlapping relationships involving Qatar expatriate Sifat Ullah, his girlfriend and another man led to his killing in the capital’s Madartek Adarshapara Baraitola area, the Rapid Action Battalion (RAB) has said.

Three people have so far been arrested in connection with the killing. RAB-3 arrested Al Amin Hossain Joy, 27, the prime accused, and Md Sajidul Islam Sajid, 24, while police later showed Sifat’s girlfriend, Nazifa Sheikh Neha, 23, arrested in the case, RAB-3 Commander Lt Col Fayezul Arefin said at a briefing at the force’s Shahjahanpur headquarters today (7 September).

According to RAB, Sifat had returned from Qatar around six weeks ago and had been in a relationship with Neha for a long time.

Sajid had also been in a relationship with Neha for more than four years but was unaware of her relationship with Sifat, RAB said. He learnt about it at Neha’s birthday celebration on 25 August, after which his relationship with her ended.

Joy, meanwhile, had known Neha since their school days. RAB said he was known in the area as a local “boro bhai” and alleged that he sometimes visited Neha’s room with friends to take drugs.

Sifat learnt about the visits and asked them not to come to the room, creating resentment between them, according to RAB.

Neha had rented a sublet room in Madartek Baraitola to run an online business.

At around 3pm yesterday (6 September), Sifat went there on his motorcycle to meet her. RAB said Joy’s associates Yasin, Anik, Tuhin and Raju, along with several others, later arrived at the house, while Joy remained downstairs.

An argument and scuffle broke out between Sifat and the group inside the room, leaving Yasin and Raju injured, RAB said.

The group then went downstairs, followed by Sifat, who tried to leave the area on his motorcycle.

RAB said Joy’s associates threw bricks at Sifat from behind, causing him to fall. Joy then allegedly struck him hard on the head with a stick, while others attacked him with sharp weapons. His motorcycle was also vandalised.

Local residents took the critically injured Sifat to Mugda Medical College Hospital. As his condition worsened, he was transferred to Dhaka Medical College Hospital, where he died while undergoing treatment.

Sifat’s family later filed a murder case with Sabujbagh Police Station.

RAB arrested Sajid at around 2am today from the Hajibari Balurpar area of Khilgaon. Based on information obtained from him, Joy was arrested at around 5am from the Shekher Jayga Bazar area under Khilgaon Police Station.

RAB said Joy was directly involved in Sifat’s killing, while Sajid, the second accused in the case, was arrested after his alleged involvement was identified through digital investigation. The force said they were all part of the same circle of friends.

Sabujbagh Police Station Officer-in-Charge Saiful Islam said Neha had been taken into police custody after the incident. She was later shown arrested after the case was filed and police found evidence of her alleged involvement.

Researchers at the US Department of Energy’s Princeton Plasma Physics Laboratory (PPPL) and Princeton University have developed an artificial intelligence-based software framework capable of making rapid decisions to control fusion plasma at speeds beyond human response.

The framework, called PACMAN, or prediction and control using machine learning, was successfully tested in five experiments at the DIII-D National Fusion Facility tokamak in San Diego, according to the researchers.

Their findings were published in the journal “Nuclear Fusion”.

Fusion research seeks to replicate the process that powers the sun to produce a potentially abundant source of electricity. Tokamaks use powerful magnetic fields to confine extremely hot plasma, but keeping the plasma stable requires continuous adjustments to heating systems, magnets, and gas injectors.

Plasma instabilities can develop within milliseconds, making them too fast for human operators to address. Computer simulations can predict plasma behaviour, but some simulations take days or months and therefore cannot provide real-time control during experiments.

Machine-learning models, however, can predict plasma behaviour within milliseconds, researchers said.

PACMAN integrates multiple AI models into a single control system. It collects real-time measurements, such as plasma temperature, density and magnetic signals, checks the data and uses AI models to assess the plasma’s current and predicted behaviour.

Controllers then use those predictions to determine necessary adjustments, such as changing heating power. Before commands are sent to the tokamak, PACMAN checks them against hardware safety limits and resolves conflicting instructions.

During five experiments at DIII-D, the system demonstrated several capabilities. It allowed a reinforcement-learning model to control plasma heating, predicted bursts of energy from the plasma edge, detected and controlled waves caused by fast particles, adjusted plasma density and rotation, and predicted a potentially disruptive instability known as a tearing mode.

In one experiment, the AI predicted the tearing mode about 200 milliseconds before it was expected to occur, allowing researchers to adjust the plasma to prevent the instability rather than suppressing it after it began.

PACMAN also coordinated all six of DIII-D’s gyrotrons, which heat plasma using powerful microwave beams. The system simultaneously adjusted their power and mirror positions to achieve targets set by researchers.

Researchers said the framework could significantly speed up the development and testing of new AI models. While integrating the first model took months, adding a second model took only a few days, allowing researchers to test, retrain, and replace models more quickly.

The researchers stressed that PACMAN is not designed to eliminate human oversight. Human operators continue to set the control objectives and parameters, while the framework enforces hardware safety limits regardless of AI recommendations.

Egemen Kolemen, an associate professor at Princeton University and PPPL, said PACMAN’s modular design could allow AI algorithms to be added, replaced, or operated simultaneously without requiring changes to the rest of the system.

The researchers believe the framework could eventually be adapted for tokamaks of different designs and sizes, potentially providing a common platform for AI-based plasma control across future fusion facilities.

Highlights:

• Five died when an Amazon Prime Air plane overran a runway and hit vehicles, officials say.
• Five others were hospitalised, including three in critical condition, fire chief says
• The Boeing 767-300 operated by 21 Air on behalf of Amazon departed from Puerto Rico

At least five people died when an Amazon Prime Air cargo flight overran a runway and crashed while landing at Miami International Airport on Sunday, local officials said.

Prime Air Flight 7598 overran the airport’s diagonal runway shortly before 2 pm EDT (1800 GMT) and struck multiple vehicles on the ground, officials said during a news conference on Sunday afternoon.

Five people were also injured in the crash, including three who are in critical condition, said Miami-Dade Fire Chief Ray Jadallah. At least two people were trapped in vehicles that were struck by the plane, and the pilot and co-pilot were trapped inside the plane by the crash, Jadallah said.

Officials declined to identify those who died.

More than 60 rescue units and about 200 personnel responded to the incident, Miami-Dade Fire Rescue said in a statement. As of late afternoon, emergency personnel were still dealing with an active fuel leak from the aircraft, Jadallah said.

Although the airport halted all flights immediately after the crash, two of its four runways had been reopened by Sunday evening.

The plane, which departed from San Juan, Puerto Rico, was a 32-year-old Boeing 767-300 freighter that had operated as a passenger jet for various airlines from 1994 to 2015, according to flight tracking service Flightradar24.

Cargo airline 21 Air operated the plane for Amazon, an Amazon spokesperson said in a statement.

“We are devastated by the accident involving one of our aircraft in Miami today,” 21 Air said on its website. “Our deepest condolences are with the families and loved ones of those who lost their lives.”

Reuters photos after the crash showed the jet with its nose on the ground and tail tilted up. The right side of the plane had what appeared to be scorch marks.

Flightradar24 said the plane was travelling at 112 knots (129 mph) when it exited the usable runway.

The National Transportation Safety Board, which investigates US civil aviation accidents, said it had sent a team led by Chairwoman Jennifer Homendy to investigate the accident and would hold its first media briefing in Miami on Monday.

Amazon is still gathering details about what happened, company spokesperson Kelly Nantel said. “We’re working closely with local authorities and officials to understand exactly what happened,” Nantel said in a statement. “Right now, our absolute priority is the safety, well-being, and care of everyone involved.”

Boeing and 21 Air said they would support the government investigations into the crash.

Sunday’s incident is not the first time an Amazon Prime Air plane has been involved in a fatal crash. In February 2019, Atlas Air Flight 3591, a Boeing 767-300 freighter operating on behalf of Amazon Prime Air from Miami to Houston, entered a rapid descent and crashed into Trinity Bay, Texas. All three people on board were killed.