Meta tags:

Headings (most frequently used words):

research, themes, recent, talks, teaching, intelligence, amplification, lab, ialab, publications, daniel, khashabi,

Text of the page (most frequently used words):
and (171), daniel (122), #khashabi (122), the (101), for (92), conference (67), code (66), slides (63), language (61), reasoning (56), evaluation (43), computational (40), linguistics (38), association (35), talk (34), learning (33), 2026 (33), wang (32), 2025 (32), model (30), zhang (30), arxiv (29), data (28), training (28), safety (28), van (28), processing (27), poster (27), benjamin (27), durme (27), natural (25), with (24), agents (24), 2024 (23), dan (22), roth (22), international (21), emnlp (20), empirical (19), methods (19), ashish (18), science (18), meeting (18), jingyu (18), tushar (17), khot (17), acl (17), findings (17), chapter (16), sabharwal (16), annual (16), models (16), 2023 (16), context (16), 2022 (15), preprint (14), retrieval (14), 2021 (14), machine (14), seminar (14), naacl (13), chen (13), andrew (12), north (12), american (12), mishra (12), project (12), jiang (12), can (12), systems (11), liu (11), humans (11), text (11), interpretability (11), long (11), tianjian (11), via (10), information (10), from (10), nlp (10), grounding (10), are (10), hajishirzi (10), multilingual (10), choi (10), generation (10), shen (10), self (10), research (10), advised (10), peter (9), demo (9), 2019 (9), university (9), 2020 (9), hannaneh (9), yejin (9), iclr (9), multimodal (8), workshop (8), william (8), intelligence (8), question (8), answering (8), 2018 (8), world (8), representations (8), adam (8), icml (8), colm (8), transactions (7), knowledge (7), artificial (7), nicholas (7), slides2 (7), more (7), lingfeng (7), anqi (7), llms (7), modeling (7), large (7), llm (7), how (7), huang (7), students (7), advances (6), neurips (6), 2016 (6), clark (6), mark (6), kyle (6), award (6), smith (6), human (6), yining (6), 2015 (5), neural (5), aaai (5), ben (5), zhou (5), song (5), your (5), michael (5), driven (5), understanding (5), sihao (5), diverse (5), richardson (5), translation (5), kevin (5), noah (5), impact (5), swaroop (5), kim (5), hou (5), aayush (5), dawn (5), lawrie (5), andrews (5), byerly (5), gao (5), speech (5), kenton (5), jurayj (5), alvin (5), tianmin (5), shu (5), hopkins (5), please (5), peng (4), scale (4), liang (4), oyvind (4), tafjord (4), answer (4), singh (4), christos (4), christodoulopoulos (4), over (4), vivek (4), open (4), multi (4), than (4), robust (4), benchmark (4), philipp (4), koehn (4), tasks (4), zhao (4), you (4), discovery (4), sean (4), welleck (4), case (4), paper (4), ximing (4), time (4), blog (4), generalization (4), tanay (4), yeganeh (4), kordi (4), trust (4), weiting (4), tan (4), improves (4), haoran (4), weller (4), marc (4), marone (4), supervised (4), yang (4), european (4), murray (4), dongwei (4), arda (4), uzunoglu (4), jiefu (4), adi (4), asija (4), sun (4), scientific (4), papers (4), ahmed (4), embodied (4), engineering (4), eric (4), publications (4), lab (4), fall (4), center (4), prospective (4), institute (4), joint (3), jeffrey (3), john (3), oren (3), etzioni (3), inference (3), questions (3), parisa (3), kordjamshidi (3), sameer (3), ijcai (3), srikumar (3), 2017 (3), statistical (3), what (3), semantic (3), luo (3), tse (3), yangqiu (3), gupta (3), qiang (3), ning (3), challenge (3), phd (3), linguistic (3), chris (3), callison (3), burch (3), perspectives (3), study (3), temporal (3), not (3), tacl (3), others (3), faeze (3), brahman (3), challenges (3), implicit (3), jungo (3), kasai (3), wei (3), chang (3), lianhui (3), qin (3), best (3), chitta (3), baral (3), instructions (3), david (3), patel (3), yizhong (3), generated (3), outstanding (3), bias (3), video (3), alex (3), when (3), non (3), tmlr (3), ali (3), andrea (3), quantifying (3), yichen (3), tianxing (3), abe (3), bohan (3), zhengping (3), orion (3), cheng (3), xiao (3), benchmarking (3), muhan (3), taiming (3), alan (3), yuille (3), jieneng (3), efficient (3), eacl (3), nathaniel (3), weir (3), presentation (3), there (3), nayak (3), kate (3), sanders (3), off (3), paul (3), behavior (3), health (3), dredze (3), sungwon (3), hannah (3), weiqi (3), kaiser (3), alignment (3), owen (3), bianchi (3), maya (3), willey (3), chelsea (3), nicole (3), kuznetsov (3), mathew (3), koretsky (3), mike (3), nalls (3), faraz (3), faghri (3), duh (3), rag (3), nalisnick (3), zhan (3), planning (3), 2604 (3), recent (3), spring (3), student (3), seminars (3), jhu (3), graduate (3), email (3), that (3), this (3), creative (3), 2014 (2), image (2), sebastian (2), nowozin (2), jeremy (2), jancsary (2), fitzgibbon (2), demosaicing (2), acquisition (2), haoruo (2), hao (2), problems (2), wieting (2), feng (2), clustering (2), google (2), better (2), programming (2), conferences (2), structured (2), resources (2), lrec (2), sammons (2), feature (2), extraction (2), relational (2), networks (2), tom (2), tsai (2), shyam (2), upadhyay (2), snigdha (2), chaturvedi (2), beyond (2), pennsylvania (2), erfan (2), sadeqi (2), azer (2), limitations (2), under (2), yin (2), claims (2), leaderboard (2), going (2), commonsense (2), coverage (2), bhavana (2), dalvi (2), carissa (2), schoenick (2), sumithra (2), bhakthavatsalam (2), all (2), supervision (2), sewon (2), min (2), unifiedqa (2), format (2), system (2), biases (2), marzieh (2), suite (2), jonathan (2), use (2), angela (2), lourie (2), pal (2), kai (2), have (2), direct (2), ai2 (2), transparent (2), web (2), cambridge (2), toward (2), prompt (2), prompts (2), liwei (2), youngjae (2), west (2), decoding (2), constrained (2), cross (2), purohit (2), jacob (2), doshi (2), kumar (2), maarten (2), based (2), aligning (2), benchmarks (2), zhong (2), parametric (2), boyuan (2), zheng (2), efficiency (2), representation (2), generating (2), correct (2), xiang (2), allen (2), amanda (2), dai (2), game (2), semstamp (2), watermark (2), transformers (2), barrier (2), identification (2), insights (2), into (2), compression (2), pre (2), spotlight (2), jin (2), consistency (2), yet (2), advanced (2), chandler (2), candice (2), morgan (2), struggle (2), many (2), creativity (2), verifiable (2), policy (2), preference (2), miriam (2), wanner (2), core (2), fang (2), icl (2), ciphers (2), evaluating (2), readability (2), good (2), user (2), walden (2), grounded (2), elgohary (2), magooda (2), feedback (2), adaptation (2), 2509 (2), jack (2), jason (2), weston (2), jointly (2), quality (2), performance (2), positional (2), shah (2), biomedical (2), era (2), eugene (2), conformal (2), yifan (2), mei (2), wen (2), big (2), mehrdad (2), farajtabar (2), risk (2), contexts (2), weak (2), rethinking (2), alvarado (2), ziyang (2), building (2), balancing (2), exploration (2), activations (2), hongyuan (2), collaborate (2), vishal (2), loop (2), shi (2), katherine (2), agent (2), 09443 (2), 2606 (2), zheyuan (2), ended (2), number (2), undergraduate (2), here (2), our (2), 601 (2), teaching (2), nyu (2), tel (2), aviv (2), johns (2), intelligent (2), department (2), apple (2), talks (2), application (2), fill (2), out (2), form (2), visiting (2), minimum (2), six (2), months (2), work (2), before (2), was (2), post (2), doc (2), aligned (2), oversight (2), mechanisms (2), autonomy (2), across (2), principles (2), these (2), themes (2), ieee, tip, denoising, learned, nonparametric, random, fields, cognitive, applications, cognitum, zhiye, fei, illinois, profiler, schemas, solving, hard, coreference, kent, quanrud, online, adversarial, delays, 1508, 06235, yufei, side, probabilistic, deterministic, algorithm, patent, 344, 690, patents, reinhard, bernhard, danyal, martin, bruce, justin, lindbloom, turney, combining, statistics, elementary, coling, bhargav, mangipudi, call, saul, flexible, integer, semi, edison, simplified, seventh, starai, summons, saurabh, sinha, querying, heterogeneous, signll, conll, essential, global, abstractions, redman, rizzolo, lev, ratinov, guanheng, quang, subhro, roy, stephen, mayhew, zhili, xiaodong, shashank, naveen, arivazhagan, shaoshi, ling, cogcompnlp, swiss, army, knife, looking, surface, set, reading, comprehension, multiple, sentences, zero, shot, entity, typing, type, compatible, thesis, possibilities, hop, imperfections, visualization, wenpeng, seeing, things, different, angle, discovering, about, demonstrations, perspectroscope, window, vacation, takes, longer, walk, magazine, niket, tandon, dirk, groeneveld, michal, guerquin, schmitz, regents, exams, overview, aristo, hongming, transomcs, graphs, slides3, sabhwawal, created, equal, choosing, right, approach, assess, hypotheses, common, sense, minimal, bang, buck, perturbation, crossing, boundaries, single, tao, unqovering, stereotypical, underspecified, arman, cohan, siamak, shakeri, pedram, hosseini, pouya, pezeshkpour, malihe, alikhani, moin, aminnaseri, bitaab, sarik, ghazarian, parsinlu, persian, mor, geva, elad, segal, berant, did, aristotle, laptop, strategies, wmt, farhad, akhbardeh, arkady, arkhangorodsky, magdalena, biesialska, ond, bojar, rajen, chatterjee, vishrav, chaudhary, marta, costa, jussa, cristina, espa, bonet, fan, christian, federmann, markus, freitag, yvette, graham, roman, grundkiewicz, barry, haddow, leonie, harter, kenneth, heafield, christopher, homan, matthias, huck, kwabena, amponsah, kaakyire, knight, kocmi, christof, monz, makoto, morishita, masaaki, nagata, ajay, nagesh, toshiaki, nakazawa, matteo, negri, santanu, allahsera, auguste, tapo, marco, turchi, valentin, vydrin, marcos, zampieri, wmt21, modular, decompose, existing, jieyu, ethical, advice, taker, understand, interventions, amos, gooaq, types, 2102, 03315, think, solved, try, arc, creating, internet, perspective, press, automatic, xinxi, lyu, waywardness, curious, discretized, interpretation, continuous, ronan, bras, rowan, zellers, neurologic, esque, lookahead, heuristics, kelvin, luu, suchin, gururangan, karishma, mandyam, waits, one, analysis, misalignment, lasting, task, crowdsourcing, hey, solve, complex, talking, reframing, instructional, gptk, gabriel, stanovsky, bragg, weld, genie, reproducible, standardized, pegah, alipoormolabashi, amirreza, mirzaei, anjana, arunkumar, arjun, ashok, arut, selvan, dhanasekaran, atharva, naik, stap, eshaan, pathak, giannis, karamanolakis, haizhi, gary, lai, ishan, ishani, mondal, anderson, kirby, kuznia, krima, maitreya, kuntal, mehrad, moradshahi, mihir, parmar, mirali, neeraj, varshney, phani, rohitha, kaza, pulkit, verma, ravsehaj, puri, rushang, karia, shailaja, keyur, sampat, savan, siddhartha, sujan, reddy, sumanta, patro, dixit, xudong, super, naturalinstructions, declarative, 1600, hyunwoo, gunhee, sap, prosocialdialog, prosocial, backbone, conversational, cold, energy, langevin, dynamics, 2202, 12359, stronger, broader, alisa, instruct, nikil, roashan, selvam, sunipa, dev, tail, wagging, dog, dataset, construction, social, mallen, rajarshi, das, victor, akari, asai, investigating, effectiveness, memories, flatness, aware, selection, accuracy, sample, anastasia, razdaibiedina, khetan, zohar, karnin, vishaal, kapoor, madan, projection, invariance, mitigates, collapse, tianxiao, sequences, finalist, certification, aarohi, srivastava, abhinav, rastogi, abhishek, rao, abu, awal, shoeb, abubakar, abid, fisch, brown, santoro, aditya, adri, garriga, alonso, agnieszka, kluska, aitor, lewkowycz, akshat, agarwal, alethea, power, ray, warstadt, alexander, kocurek, safaya, tazarv, alice, alicia, parrish, nie, aman, hussain, askell, dsouza, ambrose, slone, ameet, rahane, anantharaman, iyer, anders, andreassen, madotto, santilli, andreas, stuhlm, ller, lampinen, andy, zou, angelica, anh, vuong, animesh, anna, gottardi, imitation, extrapolating, capabilities, yung, sung, chuang, hongwei, yulia, tsvetkov, paraphrastic, robustness, pretrained, learn, gradient, descent, hanjie, rora, free, rationale, detection, yunmo, dissecting, dated, tracing, cutoffs, shreya, sharma, vijay, tiyyala, analobench, abstract, analogies, kuai, failures, know, but, don, tell, diffnorm, normalization, autoregressive, luoxin, llavolta, modal, stage, wise, visual, according, prompting, quoting, haoping, gear, augmenting, generalizable, tool, resolution, error, norm, truncation, presence, noise, linfeng, lifeng, baolin, haitao, dong, trickle, down, reward, rlhf, annals, internal, medicine, vedula, screening, search, results, systematic, reviews, journal, academics, kathryn, thompson, kimberley, emily, delinski, tutor, assessing, virtual, classroom, discriminating, responses, worldapis, worth, apis, thought, experiment, turkingbench, dixuan, upsample, upweight, balanced, heavily, imbalanced, datasets, design, quote, simplemix, frustratingly, simple, mixing, guoxuan, chuyu, rationalyst, mining, rationales, process, seth, ebner, factual, precision, scoring, informative, sub, claim, hongru, zixiao, lauren, gardner, society, generative, simulate, inform, public, vaccine, hesitancy, sophia, hager, hell, high, water, formulate, backup, plans, zhouxiang, substitution, jiacan, certified, mitigation, worst, copyright, infringement, isabel, cachola, evaluators, metrics, measures, challenging, evaluator, sycophancy, rebuttal, shaobo, seunghoon, han, may, recknor, gantt, claimcheck, critiques, xiawei, iftekhar, jackson, jailbreak, distillation, renewable, friction, fully, incorporate, external, asia, pacific, aacl, mortensen, hale, sirin, yarowsky, niyati, bafna, hypothesis, suffers, failure, controllable, requirements, genex, explorable, 02534, yiming, ping, swarnadeep, saha, lanchantin, tianlu, reinforcing, diversity, generations, 25671, flaw, averages, uniformity, 2510, 09770, gold, panning, turning, signal, document, lancet, digital, avarado, danek, khani, anant, dadu, syed, mary, makarious, cory, kristin, levine, paige, jarreau, vitale, elise, marsan, hirotaka, iwaki, hampton, leonard, sara, bandres, ciga, singleton, shekoufeh, mokhtari, cardbiomedbench, falls, short, adverse, effects, surgellm, chunsheng, zuo, embedding, domain, dense, ecir, debashish, chakraborty, principled, guarantees, prediction, maijunxian, ruisi, juyi, lin, ran, thadd, wiedemer, qingying, dezhi, yaoyao, qian, lianyu, zelong, hong, jiahui, qianli, hang, lingzi, guo, lantao, jiachen, hanwen, xing, tianqi, fengyuan, weihang, yizheng, jiao, jianheng, danyang, pengcheng, boyang, zehong, gaoyun, kitaoka, yile, hua, blacutt, tin, nguyen, siyuan, shaoyue, linyang, runming, yanzhi, mengyue, ziqiao, very, anushri, suresh, rishi, thinking, control, compute, budget, metod, jazbec, charith, peris, rinat, khaziev, wynn, controlling, corrupted, early, exiting, dayeon, marine, carpuat, mcnamee, nepotism, trading, functions, near, lossless, strong, teacher, generate, tabular, summaries, protocol, biomedsql, sql, bases, cao, shargh, jing, ruidong, mohd, zaki, wyatt, bunstine, somdatta, goswami, tyrel, mcqueen, shields, jaafar, awady, paulette, clancy, coding, reproduce, materials, jash, kuan, panoramas, hierarchies, literature, hidden, haystack, smaller, needles, difficult, find, roxana, petcu, evangelos, kanoulas, rijke, query, decomposition, exploitation, ia2, fine, tuning, haozhu, sid, amr, sharaf, mahesh, pasupuleti, waltz, oral, accepted, jiahan, muqing, nanru, shunchi, yana, jiahao, yilun, rama, chellappa, closed, zhaoyi, joey, bowei, bhiman, baghel, anneliese, brei, haw, shiuan, lorraine, meng, creativityprism, holistic, mahler, revsine, schatz, nathan, breslow, genomic, next, token, predictors, learners, 2602, 19304, haojun, suyu, guerrerio, jianzhi, chien, ming, safe, interpretable, path, cooperation, 2603, 26791, collison, cystal, characterizing, relative, scholarly, 01432, hexuan, finer, citations, always, granularity, attributed, tier, instruction, hierarchy, steered, surjective, 2605, 13625, jen, koevering, sijie, heyuan, weiyan, zhuoran, ziang, jie, interpret, 23525, chuanyang, compacting, 24893, zehao, jianwen, xie, horizon, test, continual, filter, topic, selected, also, grateful, exceptional, who, listed, fun, outing, team, photo, tajamul, ashraf, austen, liao, brian, gonzalez, nick, amplification, ialab, 771, 471, 671, show, mid, atlantic, colloquium, stanford, cmu, lti, progress, towards, holy, grail, usc, ucla, glasgow, tehran, applied, physics, umd, security, amazon, electrical, technologies, series, healthy, aging, keynote, professionals, quarterly, forum, engineered, oracle, labs, technology, lunch, mlxastro, cosmo, archaeology, ancient, history, microsoft, apply, through, list, potential, advisor, need, contact, directly, will, get, back, opportunity, fit, postdocs, above, length, visit, masters, would, like, group, commitment, hours, per, week, current, due, emails, receive, cannot, respond, every, individually, review, below, contacting, visitors, previously, working, penn, uiuc, hanna, most, following, communities, key, real, accelerating, where, grows, faster, any, individual, absorb, debate, red, teaming, verification, formalized, automated, balance, auditing, ensure, amplify, capacity, interactions, characterize, trade, between, generality, specialization, sticking, wants, adaptive, communication, interaction, fundamental, govern, emerges, guiding, scaling, foundations, pursued, vision, along, three, complementary, axes, broadly, interested, making, increasingly, engage, their, hinges, trustworthiness, agenda, lies, tension, thrives, revealing, novel, connections, deep, structural, parallels, distant, domains, inherently, prone, false, associations, undermine, rather, enhance, productivity, draws, algorithmic, tools, develop, frameworks, harness, benefits, while, remaining, values, enabling, trustworthy, reliable, helpful, danielk, edu, office, ne366, marbury, secondary, affiliations, intensive, assured, assistant, professor, computer,

Text of the page (random words):
agenda lies a tension creative reasoning thrives on revealing novel connections and deep structural parallels across distant domains yet it is inherently prone to false associations that can undermine rather than enhance human productivity my work draws on algorithmic and statistical tools to develop computational frameworks to harness the benefits of creative reasoning while remaining transparent verifiable and aligned with human values enabling trustworthy ai driven discovery i have pursued this vision along three complementary axes understanding the foundations what fundamental principles govern the intelligent behavior that emerges from large scale training and can we use these insights as guiding principles of scaling reasoning communication and interaction how can we amplify ai s capacity for reasoning across tasks contexts and interactions with the world how can we characterize the trade off between generality vs specialization sticking to what user wants in adaptive systems safety oversight and evaluation how can oversight mechanisms e g debate red teaming verification be formalized and automated at scale how can ai systems balance autonomy with auditing and safety mechanisms to ensure human trust our research is driven by a key real application accelerating scientific discovery in an era where knowledge grows faster than any individual can absorb most of my research is aligned with the following research communities natural language processing acl naacl emnlp machine learning and ai colm iclr neurips icml aaai ijcai previously i was a post doc at allen institute for ai working with yejin choi hanna hajishirzi and ashish sabharwal before my post doc i was a graduate student at penn uiuc advised by dan roth information for prospective students and visitors due to the large number of emails i receive i cannot respond to every email individually please review the information below before contacting me current jhu students if you are an undergraduate or masters student and would like to work on research with my group please fill out this form the minimum time commitment is 15 hours per week for six months prospective visiting students please fill out the above form for visiting graduate students the minimum length of a visit is six months prospective postdocs please email me directly with your cv and i will get back to you if there is an opportunity that is a good fit prospective graduate students please apply through the system and list me as a potential advisor in your application there is no need to contact me recent talks 2026 apple workshop on reasoning and planning slides 2026 microsoft slides 2026 archaeology ancient history and ai workshop slides 2025 jhu mlxastro cosmo meeting slides 2025 apple workshop on reasoning and planning slides 2025 university of pennsylvania computational linguistics lunch slides 2024 university of cambridge the language technology lab seminar slides 2024 oracle labs ml seminar slides 2024 tel aviv nlp seminar slides 2024 forum on engineered ai systems slides 2024 keynote at engineering for professionals quarterly meeting slides 2024 workshop on llms for healthy aging slides 2023 nyu text as data talk series slides 2023 hopkins center for language and speech technologies seminar video 2023 hopkins electrical engineering department seminars slides 2023 amazon human in the loop seminar 2023 hopkins center for health security seminars slides 2023 umd computational linguistics seminar slides 2023 applied physics lab intelligent systems center seminars 2022 university of tehran nlp seminar 2021 university of glasgow ir seminar slides 2021 johns hopkins university slides 2021 google ai slides 2021 ucla big data and ml seminar slides 2021 usc nlp seminar slides 2020 tel aviv university nlp seminar slides 2019 the workshop on progress towards the holy grail cp 2019 slides 2019 cmu lti seminar slides 2018 nyu nlp seminar reasoning driven question answering 2018 stanford nlp seminar slides 2018 mid atlantic student colloquium on speech language and learning slides show more teaching cs 601 471 671 nlp self supervised models spring 2023 spring 2024 spring 2025 fall 2026 cs 601 771 advances in self supervised models fall 2022 fall 2024 fall 2025 intelligence amplification lab ialab phd students adam byerly jack jingyu zhang co advised w benjamin van durme andrew wang co advised w nick andrews jiefu ou co advised w benjamin van durme tianjian li kaiser sun co advised w mark dredze hannah gonzalez co advised w benjamin van durme zheyuan brian zhang co advised w tianmin shu austen liao co advised w benjamin van durme tajamul ashraf co advised w vishal patel ziyang huang co advised w ben van durme here s a team photo from our recent fun outing we re also grateful to collaborate with a number of exceptional phd ms and undergraduate students who are not listed here publications selected recent all publications filter by topic reasoning evaluation model training safety retrieval ai humans ai for science agents interpretability grounding multimodal multilingual long context embodied 2026 open ended long horizon text game generation for test time continual learning agents agents evaluation reasoning zheyuan zhang zehao wen alvin zhang andrew wang jianwen xie daniel khashabi and tianmin shu arxiv preprint arxiv 2606 24893 2026 code project self compacting language model agents agents long context tianjian li jingyu zhang william jurayj xi wang chuanyang jin mehrdad farajtabar eric nalisnick and daniel khashabi arxiv preprint arxiv 2606 23525 2026 slides how to interpret agent behavior agents interpretability jie gao kaiser sun jen tse huang katherine van koevering sijie ji heyuan huang weiyan shi zhuoran lu ziang xiao daniel khashabi and mark dredze arxiv preprint arxiv 2605 13625 2026 steered llm activations are non surjective interpretability safety aayush mishra daniel khashabi and anqi liu arxiv preprint arxiv 2604 09443 2026 code many tier instruction hierarchy in llm agents agents safety jingyu zhang tianjian li william jurayj hongyuan zhan benjamin van durme and daniel khashabi arxiv preprint arxiv 2604 09443 2026 project are finer citations always better rethinking granularity for attributed generation grounding hexuan wang jingyu zhang benjamin van durme and daniel khashabi arxiv preprint arxiv 2604 01432 2026 code cystal characterizing relative impact of scholarly publications ai for science evaluation hannah collison benjamin van durme and daniel khashabi arxiv preprint arxiv 2603 26791 2026 safe and interpretable multimodal path planning for multi agent cooperation embodied agents multimodal safety haojun shi suyu ye katherine m guerrerio jianzhi shen yifan yin daniel khashabi chien ming huang and tianmin shu arxiv preprint arxiv 2602 19304 2026 project genomic next token predictors are in context learners interpretability nathan breslow aayush mishra mahler revsine michael c schatz anqi liu and daniel khashabi transactions on machine learning research tmlr 2026 talk poster code creativityprism a holistic benchmark for large language model creativity evaluation zhaoyi joey hou bowei alvin zhang yining lu bhiman kumar baghel anneliese brei ximing lu meng jiang faeze brahman snigdha chaturvedi haw shiuan chang daniel khashabi and xiang lorraine li transactions on machine learning research tmlr 2026 data world in world world models in a closed loop world embodied multimodal jiahan zhang muqing jiang nanru dai taiming lu arda uzunoglu shunchi zhang yana wei jiahao wang vishal m patel paul pu liang daniel khashabi cheng peng rama chellappa tianmin shu alan yuille yilun du and jieneng chen international conference on learning representations iclr 2026 oral presentation 6 of accepted papers talk project the alignment waltz jointly training agents to collaborate for safety safety agents model training jingyu zhang haozhu wang eric michael smith sid wang amr sharaf mahesh pasupuleti benjamin van durme daniel khashabi jason weston and hongyuan zhan international conference on learning representations iclr 2026 ia2 alignment with icl activations improves supervised fine tuning safety interpretability model training aayush mishra daniel khashabi and anqi liu international conference on learning representations iclr 2026 code query decomposition for rag balancing exploration exploitation retrieval reasoning roxana petcu kenton murray daniel khashabi evangelos kanoulas maarten de rijke dawn lawrie and kevin duh conference of the european chapter of the association for computational linguistics eacl 2026 hidden in the haystack smaller needles are more difficult for llms to find evaluation long context owen bianchi mathew j koretsky maya willey chelsea x alvarado tanay nayak adi asija nicole kuznetsov mike a nalls faraz faghri and daniel khashabi conference on language modeling colm 2026 code from papers to panoramas building hierarchies of scientific literature at scale ai for science evaluation muhan gao jash shah weiqi wang kuan hao huang and daniel khashabi conference on language modeling colm 2026 data can coding agents reproduce findings in computational materials science agents ai for science evaluation ziyang huang yi cao ali k shargh jing luo ruidong mei mohd zaki zhan liu wyatt bunstine william jurayj somdatta goswami tyrel mcqueen michael shields jaafar el awady paulette clancy benjamin van durme nicholas andrews william walden and daniel khashabi conference on language modeling colm 2026 data code project biomedsql text to sql for scientific reasoning on biomedical knowledge bases ai for science reasoning evaluation mathew j koretsky maya willey adi asija owen bianchi chelsea x alvarado tanay nayak nicole kuznetsov sungwon kim mike a nalls daniel khashabi and faraz faghri conference on language modeling colm 2026 data code can llms generate tabular summaries of science papers rethinking the evaluation protocol ai for science evaluation weiqi wang jiefu ou yangqiu song benjamin van durme and daniel khashabi annual meeting of the association for computational linguistics acl 2026 data slides slides2 trust functions near lossless weak to strong generalization by learning when to trust the weak teacher model training safety arda uzunoglu alvin zhang and daniel khashabi international conference on machine learning icml 2026 linguistic nepotism trading off quality for language preference in multilingual rag multilingual retrieval dayeon ki marine carpuat paul mcnamee daniel khashabi eugene yang dawn lawrie and kevin duh international conference on machine learning icml 2026 spotlight presentation code controlling the risk of corrupted contexts for language models via early exiting safety andrea wynn metod jazbec charith peris rinat khaziev anqi liu daniel khashabi and eric nalisnick international conference on machine learning icml 2026 poster code conformal thinking risk control for reasoning on a compute budget reasoning xi wang anushri suresh alvin zhang rishi more william jurayj benjamin van durme mehrdad farajtabar daniel khashabi and eric nalisnick international conference on machine learning icml 2026 poster a very big video reasoning suite multimodal reasoning evaluation maijunxian wang ruisi wang juyi lin ran ji thadd a us wiedemer qingying gao dezhi luo yaoyao qian lianyu huang zelong hong jiahui ge qianli ma hang he yifan zhou lingzi guo lantao mei jiachen li hanwen xing tianqi zhao fengyuan yu weihang xiao yizheng jiao jianheng hou danyang zhang pengcheng xu boyang zhong zehong zhao gaoyun fang john kitaoka yile xu hua xu kenton blacutt tin nguyen siyuan song haoran sun shaoyue wen linyang he runming wang yanzhi wang mengyue yang ziqiao ma and and others international conference on machine learning icml 2026 project principled context engineering for rag statistical guarantees via conformal prediction retrieval debashish chakraborty eugene yang daniel khashabi dawn lawrie and kevin duh european conference on information retrieval ecir 2026 more than efficiency embedding compression improves domain adaptation in dense retrieval retrieval model training chunsheng zuo and daniel khashabi surgellm workshop structured understanding retrieval and generation in llms era 2026 best paper award self consistency falls short the adverse effects of positional bias on long context problems reasoning long context adam byerly and daniel khashabi transactions of the association for computational linguistics tacl 2026 cardbiomedbench a benchmark for evaluating large language model performance in biomedical research ai for science evaluation owen bianchi maya willey chelsea x avarado benjamin danek marzieh khani nicole kuznetsov anant dadu syed shah mathew j koretsky mary b makarious cory weller kristin s levine sungwon kim paige jarreau dan vitale elise marsan hirotaka iwaki hampton leonard sara bandres ciga andrew b singleton mike a nalls shekoufeh mokhtari daniel khashabi and faraz faghri the lancet digital health 2026 data code blog 2025 gold panning turning positional bias into signal for multi document llm reasoning reasoning retrieval long context adam byerly and daniel khashabi arxiv preprint arxiv 2510 09770 2025 the flaw of averages quantifying uniformity of performance on benchmarks evaluation arda uzunoglu tianjian li and daniel khashabi arxiv preprint arxiv 2509 25671 2025 jointly reinforcing diversity and quality in language model generations safety model training tianjian li yiming zhang ping yu swarnadeep saha daniel khashabi jason weston jack lanchantin and tianlu wang arxiv preprint arxiv 2509 02534 2025 code genex generating an explorable world embodied multimodal taiming lu tianmin shu alan yuille daniel khashabi and jieneng chen international conference on learning representations iclr 2025 slides poster project demo blog controllable safety alignment inference time adaptation to diverse safety requirements safety jingyu zhang ahmed elgohary ahmed magooda daniel khashabi and benjamin van durme international conference on learning representations iclr 2025 data slides slides2 talk poster code the translation barrier hypothesis multilingual generation with large language models suffers from implicit translation failure multilingual niyati bafna tianjian li kenton murray david r mortensen david yarowsky hale sirin and daniel khashabi the asia pacific chapter of the association for computational linguistics aacl 2025 code feedback friction llms struggle to fully incorporate external feedback reasoning dongwei jiang alvin zhang andrew wang nicholas andrews and daniel khashabi advances in neural information processing systems neurips 2025 code jailbreak distillation renewable safety benchmarking safety evaluation jingyu zhang ahmed elgohary xiawei wang a s m iftekhar ahmed magooda benjamin van durme daniel khashabi and kyle jackson conference on empirical methods in natural language processing emnlp findings 2025 data talk claimcheck how grounded are llm critiques of scientific paper...