Language models reach 33% of 2020s entrants, and company shares run from 8% in pattern recognition to 91% in AI infrastructure

Every profile in our research carries 2 to 4 topic tags—26,140 tag assignments across 7,547 people and 5,171 distinct topics, of which 369 have at least 10 people. Across all 7,547 profiles, computer vision (1,369 people, 18%), language models (1,078 people, 14%) and machine learning (1,039 people, 14%) are the three most common tags. Grouping the 6,961 people with a dated non-education work position by the decade of their first position shows that across the seven entry decades with at least 100 people from the 1960s to the 2020s, each entry decade’s leading topic was computer vision, machine learning or language models. Language models rose from 0% of entrants in the 1960s (1 of 213) to 22% in the 2010s (385 of 1,732) and 33% in the 2020s (273 of 822).

Across all 6,766 people with a known current organization (40% at companies and 60% at universities, including 539 who hold roles in both sectors), the 60 topics with at least 50 placed people span 83 percentage points in company employment: 91% of placed people in AI infrastructure (71 of 78) and 84% in multimodal models (70 of 83) currently work at a company, compared with 8% in pattern recognition (9 of 118). Universities employ 93% of placed people in quantum information (55 of 59) and 92% in signal processing (48 of 52).

Line chart of topic shares across 7 entry decades from the 1960s to the 2020s for computer vision, machine learning, language models and reinforcement learning.

Source: People of Superintelligence career records from biographies and OpenAlex, as of 22 October 2026.

Data table

Topic	1960s (n=213)	1970s (n=457)	1980s (n=845)	1990s (n=1,230)	2000s (n=1,554)	2010s (n=1,732)	2020s (n=822)	Dated entrants
Computer vision	17% (37)	19% (86)	15% (126)	20% (244)	23% (359)	19% (321)	13% (108)	1,281
Machine learning	6% (13)	12% (54)	19% (160)	22% (271)	17% (264)	9% (158)	7% (57)	977
Language models	0% (1)	1% (6)	3% (24)	5% (63)	9% (147)	22% (385)	33% (273)	899
Robotics	11% (24)	8% (35)	8% (66)	10% (119)	7% (105)	6% (105)	4% (36)	490
Reinforcement learning	0% (1)	2% (8)	2% (19)	3% (39)	5% (83)	10% (167)	12% (99)	416
Natural language processing	8% (16)	9% (39)	7% (61)	7% (80)	5% (85)	5% (85)	4% (29)	395
Deep learning	2% (5)	1% (6)	5% (39)	4% (49)	6% (90)	5% (79)	3% (22)	290
AI policy	1% (2)	2% (7)	2% (19)	4% (49)	5% (70)	5% (95)	5% (43)	285
Neural networks	10% (22)	7% (33)	5% (46)	4% (50)	2% (33)	1% (24)	1% (9)	217
AI safety	0% (0)	0% (2)	1% (10)	1% (14)	3% (42)	5% (82)	7% (57)	207
Medical imaging	2% (5)	2% (11)	3% (27)	4% (47)	4% (55)	3% (48)	1% (12)	205
Speech recognition	6% (12)	4% (19)	5% (39)	3% (33)	1% (23)	2% (29)	1% (11)	166

Which topics grew and shrank across entry decades

We place 6,853 people across the seven entry decades from the 1960s to the 2020s, with the largest cohort in the 2010s at 1,732 entrants (another 108 people have work records starting before 1960, and 586 profiles have no dated work start). In the 1960s (213 entrants), computer vision led at 17% (37 people, including Dwijesh Dutta Majumder, Ernst Dickmanns and Leonard Uhr), followed by pattern recognition (13%, 27, including Igor Aleksander, Toshiyuki Sakai and Woody Bledsoe), robotics (11%, 24) and neural networks (10%, 22). In the 1970s (457 entrants), computer vision led at 19% (86 people), followed by machine learning (12%, 54), natural language processing (9%, 39) and pattern recognition (8%, 36).

In the 1980s (845 entrants), machine learning led at 19% (160 people, including Andrew Knyazev, Bo Zhang and Chin-Hui Lee), followed by computer vision (15%, 126), robotics (8%, 66) and natural language processing (7%, 61). In the 1990s (1,230 entrants), machine learning led at 22% (271 people), followed by computer vision (20%, 244), robotics (10%, 119) and natural language processing (7%, 80). In the 2000s (1,554 entrants), computer vision led at 23% (359 people, including Antonio Torralba, Ba‐Ngu Vo and Estibaliz Garrote), followed by machine learning (17%, 264), language models (9%, 147) and robotics (7%, 105).

In the 2010s (1,732 entrants), language models led at 22% (385 people, including Chen Lin, Endang Aminudin Aziz and Fraser Kelton), followed by computer vision (19%, 321), reinforcement learning (10%, 167) and machine learning (9%, 158). In the 2020s (822 entrants), language models led at 33% (273 people, including An Yang, Andrew Zhao and Aston Zhang), followed by computer vision (13%, 108, including Adrien Bardes, Alice Xiang and Angela Dai), reinforcement learning (12%, 99), AI safety and machine learning (7%, 57 each) and AI policy (5%, 43).

Diverging bar chart comparing percentage-point changes in topic share between 1960s–2000s entrants and 2010s–2020s entrants across 16 topics.

Source: People of Superintelligence career records from biographies and OpenAlex, as of 22 October 2026.

Data table

Topic	1960s–2000s entrants	1960s–2000s share	2010s–2020s entrants	2010s–2020s share	Change (points)	1960s share	2020s share
Language models	241	6%	658	26%	+20	0%	33%
Reinforcement learning	150	3%	266	10%	+7	0%	12%
AI safety	68	2%	139	5%	+3	0%	7%
Generative models	44	1%	99	4%	+3	0%	3%
AI policy	147	3%	138	5%	+2	1%	5%
Robot learning	49	1%	70	3%	+2	0%	2%
Multimodal AI	38	1%	75	3%	+2	0%	4%
Multimodal models	18	0%	55	2%	+2	0%	3%
Vision-language models	15	0%	41	2%	+2	0%	2%
AI alignment	10	0%	45	2%	+2	0%	2%
AI for science	67	2%	77	3%	+1	0%	3%
AI governance	53	1%	61	2%	+1	0%	3%
Optimization	60	1%	50	2%	+1	0%	2%
Generative AI	52	1%	54	2%	+1	0%	2%
AI agents	40	1%	53	2%	+1	0%	3%
Embodied AI	29	1%	46	2%	+1	0%	2%
Formal mathematics	27	1%	44	2%	+1	0%	2%
Diffusion models	22	1%	45	2%	+1	0%	3%
Multimodal learning	19	0%	38	1%	+1	0%	1%
Deep learning	189	4%	101	4%	0	2%	3%
AI ethics	67	2%	51	2%	0	1%	1%
Explainable AI	55	1%	33	1%	0	0%	2%
Human-robot interaction	59	1%	22	1%	0	0%	0%
Human-computer interaction	60	1%	17	1%	0	2%	0%
Large language models	41	1%	31	1%	0	0%	1%
AI infrastructure	34	1%	37	1%	0	0%	2%
Autonomous driving	34	1%	29	1%	0	0%	1%
Responsible AI	28	1%	35	1%	0	0%	1%
AI security	27	1%	33	1%	0	0%	1%
Computational biology	41	1%	17	1%	0	0%	0%
AI education	27	1%	29	1%	0	0%	2%
Computational neuroscience	38	1%	18	1%	0	3%	1%
Computer graphics	39	1%	16	1%	0	0%	1%
Healthcare AI	31	1%	24	1%	0	0%	1%
Medical AI	36	1%	16	1%	0	0%	0%
AI evaluation	22	1%	29	1%	0	0%	2%
Medical imaging	145	3%	60	2%	-1	2%	1%
Speech recognition	126	3%	40	2%	-1	6%	1%
Quantum computing	94	2%	22	1%	-1	1%	1%
Machine translation	73	2%	31	1%	-1	2%	1%
Multi-agent systems	77	2%	24	1%	-1	2%	1%
Information retrieval	74	2%	23	1%	-1	2%	1%
Image processing	63	1%	4	0%	-1	4%	0%
Biometrics	57	1%	4	0%	-1	0%	0%
Quantum information	55	1%	4	0%	-1	1%	0%
Evolutionary computation	52	1%	4	0%	-1	0%	0%
Signal processing	49	1%	4	0%	-1	0%	0%
Cognitive science	42	1%	9	0%	-1	6%	1%
Question answering	41	1%	10	0%	-1	1%	0%
Speech processing	47	1%	4	0%	-1	2%	0%
Speech synthesis	42	1%	8	0%	-1	3%	0%
Robotics	349	8%	141	6%	-2	11%	4%
Knowledge representation	107	2%	8	0%	-2	4%	0%
Computational linguistics	103	2%	6	0%	-2	5%	0%
Data mining	78	2%	6	0%	-2	1%	0%
Computer vision	852	20%	429	17%	-3	17%	13%
Natural language processing	281	7%	114	4%	-3	8%	4%
Neural networks	184	4%	33	1%	-3	10%	1%
Artificial intelligence	108	3%	10	0%	-3	9%	0%
Pattern recognition	154	4%	0	0%	-4	13%	0%
Machine learning	762	18%	215	8%	-10	6%	7%

Comparing the 2,554 people who entered in the 2010s–2020s with the 4,299 who entered in the 1960s–2000s across the 61 topics with at least 50 dated entrants, language models grew the most (+20 percentage points, from 6% to 26%), followed by reinforcement learning (+7 points, from 3% to 10%), AI safety (+3 points, from 2% to 5%) and generative models (+3 points, from 1% to 4%). In the other direction, machine learning shrank the most between the two periods (−10 percentage points, from 18% in the 1960s–2000s to 8% in the 2010s–2020s), followed by pattern recognition (−4 points, from 4% to 0%), computer vision (−3 points, from 20% to 17%), natural language processing (−3 points, from 7% to 4%), neural networks (−3 points, from 4% to 1%) and artificial intelligence (−3 points, from 3% to 0%). Measured from the 1960s to the 2020s, pattern recognition (−13 points, from 13% to 0%) and neural networks (from 10% to 1%) and artificial intelligence (from 9% to 0%), each −9 points, had the largest decreases, and machine learning fell 15 points from its 1990s peak of 22% to 7% in the 2020s.

Which topics are most concentrated at companies versus universities

Across all 6,766 people with at least one known current parent organization, 40% (2,692) hold a company role, 60% (4,047) hold a university role, and 22% (1,466) hold a role at a nonprofit institute, standalone lab or government body. Those groups overlap in 539 people who hold concurrent company and university positions.

Stacked horizontal bars showing the current employer distribution across companies, dual company-university roles, institutes and universities for 14 topics.

Source: People of Superintelligence career records from biographies and OpenAlex, as of 22 October 2026.

Data table

Topic	Placed people	At companies	Company share	At universities	University share	Both company and university	At institutes, labs and government	Institute share
AI infrastructure	78	71	91%	8	10%	6	8	10%
Multimodal models	83	70	84%	17	20%	9	7	8%
Multimodal AI	119	95	80%	28	24%	8	13	11%
AI agents	99	73	74%	24	24%	9	16	16%
Language models	986	688	70%	295	30%	68	136	14%
Generative models	154	107	69%	69	45%	28	31	20%
Diffusion models	74	49	66%	25	34%	5	13	18%
Autonomous driving	59	39	66%	31	53%	12	7	12%
AI alignment	59	38	64%	17	29%	9	19	32%
Reinforcement learning	449	287	64%	197	44%	51	68	15%
Embodied AI	78	48	62%	33	42%	9	10	13%
Generative AI	110	66	60%	50	45%	11	15	14%
Machine translation	96	56	58%	44	46%	7	8	8%
Responsible AI	66	37	56%	32	48%	9	16	24%
Computer graphics	55	30	55%	37	67%	12	4	7%
Vision-language models	65	35	54%	18	28%	5	19	29%
Speech recognition	160	83	52%	85	53%	12	15	9%
AI education	57	29	51%	29	51%	3	7	12%
Large language models	75	38	51%	39	52%	5	10	13%
AI evaluation	52	26	50%	20	38%	4	16	31%
Robot learning	121	59	49%	74	61%	21	22	18%
Multimodal learning	64	31	48%	38	59%	10	13	20%
AI safety	215	99	46%	59	27%	7	89	41%
Robotics	466	206	44%	295	63%	57	74	16%
Information retrieval	96	41	43%	54	56%	6	17	18%
Computer vision	1,241	520	42%	787	63%	131	207	17%
Deep learning	294	122	41%	204	69%	39	61	21%
Formal mathematics	80	33	41%	36	45%	4	23	29%
AI security	64	26	41%	44	69%	9	10	16%
AI for science	153	62	41%	90	59%	13	37	24%
Natural language processing	369	137	37%	237	64%	31	68	18%
Optimization	110	40	36%	74	67%	9	22	20%
AI governance	116	40	34%	44	38%	7	58	50%
Graph neural networks	50	17	34%	36	72%	4	12	24%
Medical AI	53	16	30%	41	77%	5	11	21%
Computational biology	61	18	30%	47	77%	7	22	36%
AI policy	283	81	29%	85	30%	11	177	63%
Quantum computing	114	32	28%	84	74%	9	30	26%
Machine learning	982	271	28%	754	77%	88	205	21%
Healthcare AI	56	15	27%	40	71%	2	13	23%
Human-computer interaction	74	18	24%	57	77%	6	14	19%
Speech processing	52	12	23%	40	77%	2	4	8%
Neural networks	194	44	23%	155	80%	11	31	16%
Medical imaging	204	46	23%	164	80%	14	48	24%
Multi-agent systems	101	22	22%	76	75%	4	20	20%
Human-robot interaction	82	17	21%	66	80%	10	22	27%
Evolutionary computation	55	11	20%	47	85%	4	6	11%
AI ethics	119	23	19%	90	76%	6	36	30%
Artificial intelligence	93	17	18%	73	78%	6	20	22%
Data mining	83	15	18%	68	82%	4	8	10%
Computational neuroscience	53	9	17%	45	85%	4	19	36%
Image processing	57	9	16%	50	88%	5	6	11%
Signal processing	52	8	15%	48	92%	4	6	12%
Explainable AI	95	14	15%	79	83%	7	30	32%
Knowledge representation	89	13	15%	76	85%	4	18	20%
Causal inference	55	7	13%	42	76%	2	23	42%
Computational linguistics	92	11	12%	75	82%	3	20	22%
Quantum information	59	7	12%	55	93%	4	14	24%
Biometrics	55	6	11%	49	89%	2	9	16%
Pattern recognition	118	9	8%	103	87%	3	21	18%

Among the 60 topics with at least 50 placed people, AI infrastructure is the most concentrated at companies: 91% (71 of 78) hold a current company role and 10% (8) hold a university role, led on the company side by DeepSeek (9). Multimodal models follow at 84% company share (70 of 83, with 20% at universities), and multimodal AI is next at 80% (95 of 119, led by Google with 13 and University of British Columbia, University of Southern California and University of Washington among universities with 2 each). Every one of the six most company-concentrated topics—AI infrastructure, multimodal models, multimodal AI, AI agents, language models and generative models—has a company share between 69% and 91%, followed by diffusion models (66%, 49 of 74, led by Google with 12) and autonomous driving (66%, 39 of 59, led by Google with 5).

At universities, quantum information leads with 93% of placed people (55 of 59) and 12% at companies (7), led by Massachusetts Institute of Technology, Universität Innsbruck and University of Science and Technology of China (3 each). Signal processing follows at 92% university share (48 of 52, with 15% at companies). Among the four topics with the most placed people overall, the three outside the top company-share group are computer vision (63% at universities, 787 of 1,241, and 42% at companies, 520), machine learning (77% at universities, 754 of 982, and 28% at companies, 271) and robotics (63% at universities, 295 of 466, and 44% at companies, 206).

In nonprofit institutes, standalone labs and government bodies, AI policy leads with 63% of placed people (177 of 283, compared with 29% at companies and 30% at universities), followed by AI governance at 50% (58 of 116).

Across placed people with a dated career start, 57% of 2010s–2020s entrants (1,381 of 2,419) work at companies and 39% (935) work at universities, compared with 28% at companies (1,073 of 3,901) and 75% at universities (2,941) among 1960s–2000s entrants. Among 1960s–2000s entrants, company shares are 50% in language models (118 of 237) and 41% in reinforcement learning (59 of 143), compared with 31% in computer vision (244 of 780) and 23% in machine learning (169 of 728). Among 2010s–2020s entrants, company shares are 76% in language models (464 of 609) and 72% in reinforcement learning (180 of 249), compared with 62% in computer vision (249 of 399) and 41% in machine learning (86 of 210). Across primary profile categories, the leading topic is computer vision among researchers (20%, 1,276 of 6,357), language models among builders (17%, 75 of 437), language models among lab leaders (41%, 129 of 312) and AI policy among public voices (50%, 220 of 441).

Key findings

• Across 6,853 people with dated work positions in the 1960s through the 2020s, each entry decade’s leading topic was computer vision, machine learning or language models: computer vision in the 1960s (17%), 1970s (19%) and 2000s (23%); machine learning in the 1980s (19%) and 1990s (22%); and language models in the 2010s (22%) and 2020s (33%).
• Comparing 2010s–2020s entrants (2,554 people) with 1960s–2000s entrants (4,299 people) across 61 topics with at least 50 dated entrants, language models grew the most (+20 percentage points, from 6% to 26%), followed by reinforcement learning (+7 points).
• Over the same period comparison, machine learning shrank the most (−10 percentage points, from 18% to 8%), and from the 1960s to the 2020s pattern recognition (−13 points) and neural networks and artificial intelligence (−9 points) had the largest decreases.
• Among 60 topics with at least 50 placed people today, company shares are highest in AI infrastructure (91%, 71 of 78), multimodal models (84%) and multimodal AI (80%), and lowest in pattern recognition (8%).
• University shares are highest in quantum information (93%, 55 of 59), followed by signal processing (92%); nonprofit institutes, standalone labs and government bodies employ 63% of placed people in AI policy and 50% in AI governance.
• Across placed people with a dated career start, 57% of 2010s–2020s entrants work at companies and 39% work at universities, compared with 28% at companies and 75% at universities among 1960s–2000s entrants; company shares are 50% (1960s–2000s) and 76% (2010s–2020s) in language models, compared with 31% and 62% in computer vision and 23% and 41% in machine learning.