2024-05-13T17:50:39-05:00 2023-12-31 331197384 ANDERSEN TAX LLC 333 BUSH STREET SUITE 1700 SAN FRANCISCO CA 94104 990PF 2023-01-01 263115543 HARIK SHAZEER LABS HARI 2138918675 555 BRYANT STREET 513 PALO ALTO CA 94301 GEORGES HARIK CHIEF FINANCIAL OFFICER 2138918675 2024-05-13 1 GEORGES HARIK 999009999 0 MICHELLE BASS P01501825 4157642700 2 AAL1 IAL1 2023 2023-04-26 12:10:37Z X 362161 X X 844 844 844 0 0 844 844 844 0 0 0 0 120659 0 0 120659 11730 0 0 11730 9445 0 0 9445 2657 0 0 2657 144491 0 0 144491 0 0 144491 0 0 144491 -143647 844 844 502124 362161 362161 15225 15225 502124 362161 362161 2535 4939 1280 2535 6219 X 0 0 0 0 499589 355942 499589 355942 502124 362161 499589 -143647 0 355942 0 355942 12 0 12 0 12 3229 0 0 0 3229 0 3217 3217 0 0 0 0 0 0 0 0 0 0 0 1 1 CA 1 1 0 0 0 0 HTTPS://SITES.GOOGLE.COM/VIEW/WWW-HSLABS-ORG/HOME DEBORAH MCGEE ANDERSEN TAX LLC 2135932300 400 SOUTH HOPE STREET SUITE 2000 LOS ANGELES CA 90071 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 NOAM SHAZEER 555 BRYANT STREET 513 PALO ALTO CA 94301 PRESIDENT AND DIRECTOR 1.00 0 0 0 GEORGES HARIK 555 BRYANT STREET 513 PALO ALTO CA 94301 SECRETARY, CFO, AND DIRECT 4.00 0 0 0 SERGEY PANKOV 443 VENTURA AVE 5 PALO ALTO CA 94306 RESEARCHER 40.00 120659 0 0 0 NONE 0 HARIK SHAZEER LABS (THE "FOUNDATION") IS A CALIFORNIA NONPROFIT PUBLIC BENEFIT CORPORATION, ORGANIZED AND OPERATED EXCLUSIVELY FOR CHARITABLE, SCIENTIFIC AND EDUCATIONAL PURPOSES. THE PRIMARY PURPOSES OF THE FOUNDATION ARE TO PERFORM FUNDAMENTAL RESEARCH IN THE FIELD OF ARTIFICIAL INTELLIGENCE, INCLUDING BY CONVENING CONFERENCES OF EXPERTS, AND TO MAKE THE RESULTS OF SUCH RESEARCH AVAILABLE TO THE PUBLIC ON A NONDISCRIMINATORY BASIS. RESEARCH RESULTS ARE MADE PUBLICLY AVAILABLE ON THE FOUNDATION'S WEBSITE, HTTPS://SITES.GOOGLE.COM/VIEW/WWW-HSLABS-ORG/HOME, AND ALL PROGRAMMING CODE ARE MADE AVAILABLE UNDER A COMMONLY-USED OPEN SOURCE LICENSE AND WILL BE FREELY AVAILABLE ON THE INTERNET. IN 2023, THE LAB CONTINUED ON ITS MISSION TO DO ARTIFICIAL INTELLIGENCE RESEARCH AND DISSEMINATE IT WIDELY. PLEASE SEE FURTHER EXPLANATION UNDER "GENERAL EXPLANATION" AT STATEMENT 7 144491 0 0 442522 0 442522 0 0 442522 6638 435884 21794 X 144491 0 144491 X 844 740 72 440 2096 717 629 61 374 1782 144491 142640 143063 138392 568586 0 0 0 0 0 144491 142640 143063 138392 568586 0 0 14529 19566 24116 28573 86784 0 0 0 0 NOAM SHAZEER GEORGES HARIK X 0 0 14 844 0 844 0 844 0 0 0 0 0 0 0 0 0 0 ACCOUNTING FEES 11730 0 0 11730 A SUMMARY OF DIRECT CHARITABLE ACTIVITIES PART VIII-A CONTINUED EXPLANATION FROM PART VIII-A - SEE STATEMENT 5SERGEY PANKOV:DURING THE YEAR 2023 I HAVE BEEN WORKING ON THE PROBLEM OF SELF-IMPROVEMENT OF COMMON SENSE REASONING IN LANGUAGE MODELS. IN THE CONSIDERED APPROACH, A GENERATOR AND DISCRIMINATOR LANGUAGE MODEL ARE USED TO GENERATE AND EVALUATE ARGUMENTS IN SUPPORT OF PARTICULAR OUTCOMES OF BINARY DECISION PROBLEMS, WHICH ARE AUTOMATICALLY CONSTRUCTED OUT OF TEXT PASSAGES (E.G. FROM BOOKS). THE EVALUATIONS OF THE PROPOSED ARGUMENTS ARE USED TO SELECT PROMISING PROBLEMS (THAT ARE BOTH MEANINGFUL AND SOLVABLE), WHICH IN TURN CAN BE USED TO IMPROVE THE GENERATOR AND DISCRIMINATOR IN AN ITERATIVE MANNER.IT WAS ESTABLISHED THAT: A) THE FRACTION OF MEANINGFUL AND SOLVABLE PROBLEMS AMONG THE AUTOMATICALLY CONSTRUCTED PROBLEMS IS REASONABLE LARGE, B) THE GENERATED ARGUMENTS DO IMPROVE THE DISCRIMINATOR'S DECISION MAKING (RELATIVE TO THE BASELINE CASE OF NO ARGUMENT), C) THE IMPROVEMENT IS LARGER FOR LARGER MODELS (THE MODELS RANGED IN SIZE FROM HUNDREDS OF MILLIONS TO SEVERAL BILLIONS PARAMETERS).SO FAR, ONLY THE DISCRIMINATOR HAS BEEN TRAINED. THE NEXT STEP WOULD BE TO TRAIN THE GENERATOR USING DIRECT PREFERENCE OPTIMIZATION OR A SIMILAR TECHNIQUE, TO SEE IF THE ITERATIVE PROGRESS IN DECISION MAKING (AND HENCE IN COMMON SENSE REASONING) CAN BE ACHIEVED.ALSO, MY PREVIOUS WORK ON 'ENERGY-CONSERVING INTERMITTENT-CONTACT MOTION IN COMPLEX MODELS' WAS PUBLISHED IN COMMUNICATIONS IN NONLINEAR SCIENCE AND NUMERICAL SIMULATION (CNSNS 132 (2024), 107895). THIS WORK IS IMPORTANT FOR ROBOTIC LOCOMOTION, BECAUSE IT SHOWS THE WAY FOR CREATION OF ULTRA-EFFICIENT WALKING ROBOTS. OUTSIDE SERVICES - ADP 2657 0 0 2657 PAYROLL TAXES 9420 0 0 9420 STATE TAXES 25 0 0 25