←Search results

Search “OpenAI”

By relevance · Total 1248 results Clear search

Wednesday

  4. English text is being prepared.

    OpenAI launches chat lock feature for ChatGPT, protecting sensitive conversations via PIN code and biometrics

    OpenAI has launched a chat lock feature for ChatGPT. Users can set a four-digit PIN code, and on supported devices can also use Face ID or Touch ID to protect sensitive conversations. Locked conversations are moved to a separate section, isolated from regular chat history, pinned conversations, and search results, and have a separate memory space when the memory feature is enabled. The feature is rolling out gradually to Free, Go, Plus, Pro, Business, Enterprise, and Edu users, but it does not enable end-to-end encryption and does not change data retention periods.

    OpenAI has launched a chat lock feature for ChatGPT. Users can set a four-digit PIN code, and on supported devices can also use Face ID or Touch ID to protect sensitive conversations. Locked conversations are moved to a separate section, isolated from regular chat history, pinned conversations, and search results, and have a separate memory space when the memory feature is enabled. The feature is rolling out gradually to Free, Go, Plus, Pro, Business, Enterprise, and Edu users, but it does not e

    Matched fieldsTitle

    #analysis#Product updates#OpenAI

Tuesday

  1. OpenAI: Alignment Research Blog

    OpenAI proposes complementary monitorability metrics beyond g-mean²

    OpenAI proposes two complementary metrics beyond g-mean²: the pair-matched metric compares monitorability only on instances jointly supported by multiple models, while the post-control risk metric simulates the probability that the system still produces harmful outputs after the monitor rejects outputs. g-mean² was previously used to measure a monitor's ability to distinguish positive from negative samples, but because different models have different subsets of valid positive tasks, scores may reflect different task mixtures. The article notes that if the monitor is used as a deployment control measure, the model can retry after being rejected, and the risk depends on the frequency of harmful behavior, which outputs the monitor accepts, and the number of allowed attempts.

    Pair matching isolates monitorability from capability-driven instance selection

    OpenAI proposes two complementary metrics beyond g-mean²: the pair-matched metric compares monitorability only on instances jointly supported by multiple models, while the post-control risk metric simulates the probability that the system still produces harmful outputs after the monitor rejects outputs. g-mean² was previously used to measure a monitor's ability to distinguish positive from negative samples, but because different models have different subsets of valid positive tasks, scores may r

    Matched fieldsTitle

    #lab#Papers/Research#Safety / alignment#OpenAI
  4. Solidot

    The Saudi engineer suspected of spying on an OpenAI executive had a folder named Saudi Gov Docs on his phone

    Abdulrahman Alfozan, the Saudi engineer suspected of spying on an OpenAI executive, has been denied release, and his iPhone contained a folder named Saudi Gov Docs. U.S. prosecutors say his motive may have been personal business gain or acting in the interests of the Kingdom of Saudi Arabia, and noted that he had maintained contact with members of the Saudi royal family via WhatsApp. The FBI arrested him just before he was about to board a flight to leave the United States, and the one-way tickets to Riyadh for him, his wife, and their child are believed to have been paid for by the Saudi government.

    Abdulrahman Alfozan, the Saudi engineer suspected of spying on an OpenAI executive, has been denied release, and his iPhone contained a folder named Saudi Gov Docs. U.S. prosecutors say his motive may have been personal business gain or acting in the interests of the Kingdom of Saudi Arabia, and noted that he had maintained contact with members of the Saudi royal family via WhatsApp. The FBI arrested him just before he was about to board a flight to leave the United States, and the one-way ticke

    Matched fieldsTitle

    #hot-signal#Safety / alignment#OpenAI
  5. China.com Technology

    OpenAI plans to raise $30 billion, targeting a $1.4 trillion valuation

    According to Reuters citing people familiar with the matter, OpenAI is accelerating its fundraising and plans to raise about $30 billion from investors, with its valuation expected to exceed $1.4 trillion, with SoftBank Group and MGX participating. OpenAI is pitching investors on $50 billion in annualized revenue and expects annualized revenue to reach $70 billion by year-end as the GPT-6 Astra model launches. The fundraising process is intended to lock in subscription commitments before Anthropic's expected mid-November IPO; OpenAI's valuation reached $852 billion after raising $122 billion in March this year.

    According to Reuters citing people familiar with the matter, OpenAI is accelerating its fundraising and plans to raise about $30 billion from investors, with its valuation expected to exceed $1.4 trillion, with SoftBank Group and MGX participating. OpenAI is pitching investors on $50 billion in annualized revenue and expects annualized revenue to reach $70 billion by year-end as the GPT-6 Astra model launches. The fundraising process is intended to lock in subscription commitments before Anthrop

    Matched fieldsTitle

    #hot-signal#Industry news#OpenAI#Anthropic
  6. ThurrottSelected

    OpenAI Releases GPT-6 Astra, a Coding Model for Legacy Languages Such as COBOL

    OpenAI has released GPT-6 Astra, a coding model for legacy programming languages such as COBOL, Fortran, and Visual Basic. OpenAI built two evaluations for this purpose, MultiLangBench and COBOL migration. GPT-6 Astra scored 87.4% on the 20 languages in MultiLangBench, compared with 26.8% for GPT-5.6 Sol; it scored 86.8% on COBOL migration, or 97.9% at high reasoning effort, while GPT-5.6 Sol scored only 5.4%. The model is now available to developers through the OpenAI API for coding tasks in COBOL, Fortran, OCam, and Visual Basic.

    OpenAI has released GPT-6 Astra, a coding model for legacy programming languages such as COBOL, Fortran, and Visual Basic. OpenAI built two evaluations for this purpose, MultiLangBench and COBOL migration. GPT-6 Astra scored 87.4% on the 20 languages in MultiLangBench, compared with 26.8% for GPT-5.6 Sol; it scored 86.8% on COBOL migration, or 97.9% at high reasoning effort, while GPT-5.6 Sol scored only 5.4%. The model is now available to developers through the OpenAI API for coding tasks in CO

    Matched fieldsTitle

    #hot-signal#Model releases#English text is being prepared.#OpenAI
  11. Rohan Paul

    US prosecutors say former employee accused of planting eavesdropping devices in OpenAI executives' meeting rooms had Saudi government support

    BBC reports that US prosecutors say Alfozan, a former OpenAI engineer accused of planting eavesdropping devices in OpenAI executives' meeting rooms, has Saudi government ties. Prosecutors say an unnamed member of the Saudi royal family communicated with him frequently via WhatsApp and tried to recruit him to join a Saudi AI program; he is also alleged to have said in May that Saudi Arabia needed him more than the United States did, but that he would wait until his OpenAI stock vested before acting. Prosecutors say he used his wife's ChatGPT account to research making and encrypting recording devices, and used OpenAI's internal productivity agent to find rooms used by executives and external partners; OpenAI said it found no signs that model weights, source code, or customer data had been stolen.

    BBC reports that US prosecutors say Alfozan, a former OpenAI engineer accused of planting eavesdropping devices in OpenAI executives' meeting rooms, has Saudi government ties. Prosecutors say an unnamed member of the Saudi royal family communicated with him frequently via WhatsApp and tried to recruit him to join a Saudi AI program; he is also alleged to have said in May that Saudi Arabia needed him more than the United States did, but that he would wait until his OpenAI stock vested before acti

    Matched fieldsTitle

    #ai-news#Safety / alignment#OpenAI#Industry news

Monday