美國另一研究團隊以20美元開發效能超越o1-preview的AI模型
支付動態 · 2025-02-07

美國華盛頓大學與史丹佛大學的研究團隊揭露降低AI模型開發成本的新方法,以小樣本資料集與20美元硬體費用,也能訓練出推論效能超越o1-preview的AI模型

/arXiv:2501.19393

研究團隊也比較了s1-32B和DeepSeek r1和o1,測試結果並不如後二者。但研究團隊解釋,這是因為DeepSeek r1和o1都使用了大量數據進行強化學習,而s1-32B則僅使用1,000個精選資料進行微調,因此仍是最具樣本效率的模型。

本模型已開源於GitHub上。

大學研究人員不斷在以更低成本訓練高效能AI模型。今年初柏克萊大學研究團隊也用了約450美元來訓練效能近似o1-preview的模型。

Popular articles
‘A target on their back’: college athletes face wave of abuse amid gambling boom
Sports Betting
Brazil’s president says he will ban sports bets if ‘addiction’ not regulated
Sports Betting
Vietnam's tightening online gaming policy creates new market opportunities
Southeast Asia
Indiana online casino bill stalls in House committee
Regulation
Hotel-casino court rulings reveal flaws in AI price-fixing allegations
Regulation
DraftKings drops another $5M into Missouri sports betting race
Sports Betting
Major UK banks join new Gambling Harms Action Lab
Regulation
Underreported taxes on bettors’ gambling winnings leaves IRS $1.4 billion short
Sports Betting
Irish lawmakers at odds over change in gambling bill allowing ‘inducements’
Sports Betting
Netherlands goverment proses raising the minimum age for online casino slots
Regulation
NFL player calls out NFL fans upset about bets: ‘We’re human too’
Sports Betting
Italian regulator updates technical rules for gambling systems verification
Regulation
French Gambling Giant FDJ Completes €2.5bn Kindred Group Purchase
Regulation
Elevate Your Casino’s Success with Opexa Game Aggregators
HUIDU Focus
ESPN Bet collects $3M handle in New York mobile debut
Sports Betting
Home
Game
Cooperation
Find
My