I haven't read the proposal in detail but Caitlin Yardley is really talented at what she does and a pleasure to work with!

She's especially great at talking to a wide variety of creators so would be a great fit for an agency.

Description of subprojects and results, including major changes from the original proposal

Description of subprojects and results, including major changes from the original proposal

The ACX Grants 2024 award significantly strengthened our capacity for bacteriophage research and development in Nigeria. The funding allowed us to improve our laboratory infrastructure, expand our phage collection, support postgraduate research, and advance several projects from phage isolation toward characterization and potential therapeutic or agricultural applications.

One of the most important infrastructure achievements was the acquisition of an ultracentrifuge. This addressed a major limitation in our laboratory by improving our ability to concentrate and purify bacteriophages locally. Unlike expenditure on individual experiments, this is a long-term investment that will continue to support multiple phage projects, students, and collaborators beyond the grant period.

A major scientific outcome was our work on Salmonella. We isolated and characterized two novel lytic phages, Jerseyvirus ijeoma and Apdecimavirus ayanbimpe, active against Salmonella Typhi. The work progressed to peer-reviewed publication in 2026 in the study “Exploring lytic Salmonella phages as potential alternatives to antibiotics: isolation, characterization, and stability assessment of Jerseyvirus ijeoma and Apdecimavirus ayanbimpe” (Olorundare et al., 2026). The study demonstrated their lytic activity and assessed their stability under different conditions. These phages now provide candidates for the next stage of our work, particularly preclinical evaluation of safety and therapeutic efficacy.

Our work on Pseudomonas aeruginosa also expanded considerably. An MSc research project resulted in the isolation of 34 bacteriophages active against P. aeruginosa, contributing to our growing phage collection. We also completed genomic characterization of the lytic P. aeruginosa phage CSSBELLO(Igwe, Blessing Chidinma, Adesola Olaleka, and Nnaemeka Nnadi. "Isolation and Genomic Characterization of a Novel Lytic Bacteriophage (CSSBELLO) Infecting Multidrug-Resistant Pseudomonas aeruginosa, Isolated from Wastewater in Jos." Frontiera Scientia 1.1 (2026): 10-19, which was subsequently reported in a peer-reviewed publication. These studies have strengthened our ability to move from phage isolation through biological characterization to genome-level analysis. This work is also feeding into our research on a phage-based bandage for infected wounds, particularly as a potential delivery approach for difficult-to-treat wound infections.

Another important development was our expansion into phage-based vaccine research. We developed a multi-serotype Salmonella phage-based vaccine construct intended to provide broader protection across different Salmonella serotypes. Candidate components have undergone computational and experimental work, with in vivo evaluation representing the next major stage of development.

The programme also expanded beyond human health into agricultural phage biotechnology. In our Xanthomonas project, five bacterial isolates were obtained, three of which were demonstrated to be pathogenic. We subsequently isolated seven lytic bacteriophages, with three showing activity against all three pathogenic isolates tested. Plaque assays, enrichment and purification have been undertaken, and the strongest candidates provide the basis for developing a phage cocktail for future evaluation against bacterial disease in tomato. This has broadened the programme toward a One Health approach by applying the same phage expertise to both human health and food production.

We have also strengthened our capacity in phage genomics and bioinformatics. An important outcome has been the development of PhageMine, an open-source bioinformatics workflow for bacteriophage genome analysis. This development helps address one of the challenges we encountered as our phage collection expanded: the need to move beyond isolation and phenotypic characterization to systematic genomic analysis of candidate phages.

The grant has additionally supported a wider capacity-building effect. MSc and PhD students have participated in projects involving Salmonella, Pseudomonas, phage-based wound applications, and agricultural phages. The laboratory infrastructure and expertise developed through these activities are therefore supporting a growing group of researchers rather than a single project.

Major changes from the original proposal

The overall objective of building local capacity for phage research and developing phage-based solutions has remained unchanged. However, the programme became broader than originally anticipated. Our initial emphasis was largely on phage isolation, characterization, storage and therapeutic development. During implementation, this expanded to include phage-based vaccines, wound-delivery applications, agricultural biocontrol, genomic analysis, bioinformatics and postgraduate training.

We also refined our approach to clinical translation. Rather than moving directly from promising laboratory results toward clinical trials, we now recognize the importance of a staged pathway involving comprehensive genomic characterization, candidate selection, preclinical safety and efficacy studies, manufacturing and quality control, regulatory engagement and, ultimately, appropriately approved clinical evaluation.

Overall, the ACX Grants 2024 award has produced both immediate research outputs and longer-term research capacity. The acquisition of the ultracentrifuge, expansion of our phage collection, publication of characterized phages, development of new therapeutic and agricultural applications, bioinformatics capacity and training of postgraduate researchers represent outcomes that will continue beyond the grant period.

Our next phase will build on this foundation by advancing the strongest phage candidates into preclinical studies, further developing the Xanthomonas phage cocktail and phage-based bandage, strengthening the phage bank, and pursuing the manufacturing and regulatory capacity required to translate promising phages into practical applications.

I'm taking this off tomorrow if you guys won't fund it. Have a good time

@raber3 Wooo thanks for your support!

Since the original proposal, the experiments have pushed the question beyond simply asking how much information can be recovered from a transformer's own write directions. Across five models, native writes could often be followed downstream even after their original directions became difficult to recover from the accumulated residual, with their transported descendants retaining information about both their source and later functional effects. This points to a useful separation between where a computation is written, how it is transformed through the residual stream, and what part of it remains functionally relevant.

More broadly, the experiments suggest that residual perturbations can become physically high-dimensional while their behaviorally relevant structure remains comparatively low-dimensional, and that provenance and causal importance are not interchangeable. Despite the grant auto-rejection, I’m releasing these results and the associated negative findings as an extension of the original project in the hope that they’re useful for people studying circuit tracing, activation transport, and the relationship between parameter-level structure and learned representations.

Interesting project--approved!

Since publishing the ILRL Public Evaluation and Findings Project, I have created a public ILRL introduction and organized the method into a simple continuation path:

Conversation → editable review → human-AI interaction receipt → portable checkpoint → later continuation → outcome receipt

Public site: https://innovation-lineage-record-live.rrbberkovich.chatgpt.site/

I have also prepared public documentation for participation, evidence states, adaptive connections, contribution history, and correction handling. These materials are intended to support a small, privacy-safe outside-user evaluation. They do not establish independent validation yet.

The next measurable step is to invite a small number of adult participants to complete one continuation cycle. I will preserve what they received, what they understood, what confused them, what they corrected, and whether the checkpoint helped them continue later. Successful and unsuccessful outcomes will both remain part of the findings.

The Manifund grant would support the work described in the proposal: a sanitized public evaluation corpus, outside-reader sessions, adversarial cases, scoring, accessible demonstrations, privacy review, and a public findings report. Private conversations, credentials, and private lineage archives will not be published.

Current evidence state:

ESTABLISHED: the public site and core workflow exist.

READY: participant-protection and evaluation documents are being prepared.

OPEN: outside-user continuation results and Manifund funding.

NOT_MEASURED: broad usability, independent validation, market demand, and commercial value.

I welcome questions about the evaluation design, privacy boundary, scoring plan, or intended public benefit.

— Richard Ryan Berkovich

[]

BASE is doing an important work in AI Safety and needs as much support as possible.

I think you do a great work, Anna, and will be happy to donate again this year.

I think this is an interesting project -education is something that for personal and professional reasons resonates with me- and I have high trust in Jack, so I'll likely be contributing to it soon.

Tandena, I think this is great work. Thank you for doing it. I'm glad my offer to pay enticed you to do this. Post this on the EA forum?

@MarcusAbramovitch Here's the "vulture carcass disposal x human health" research proposal

Rubi was our mentor during the PRISM fellowship program when we were investigating AI preference drift during training. He gave good direction without micromanaging, which meant we could move fast and weren't dependent on him to make progress. He was also great at guiding us where to focus when needed. I have no doubt he would be great at managing a team at Principled Agents and hope to work with him again in the future.

Description of subprojects and results, including major changes from the original proposal

The project was completed and final outputs submitted to UK ARIA. The research questions I studied here formed the inspiration and basis for my continued work during the PIBBSS fellowship in 2026.

@aCFAR Not super related but wanted to say, we rented out CFAR's Bodega Bay Venue for a midpoint retreat for the Surplus incubator. It was fantastic; I'd highly encourage other folks considering doing eg a weekend retreat or workshop, to look into renting their space: https://www.rationality.org/venue

More related: many of the founders expressed an interest in participating in a future CFAR workshop. Keep up the good work!

FloraForge’s new website is now live, with an overview of the platform, current validation work, and long-term vision:

https://www.floraforgelabs.com/

Hi all — Sina here, lead architect on TROPOS-1.

A quick upfront note on the 87% Pangram flag: as an independent researcher whose primary language isn't English, I used LLMs to help polish and structure the written prose of this grant proposal.

The underlying architecture, RTL, mathematical formulation, and validation work are my own. For anyone who wants to verify rather than take that on faith:

• 87-page master dossier + executed logs: archived on Zenodo (DOI: 10.5281/zenodo.22637552)

• Synthesizable SystemVerilog RTL: passing self-checking testbenches under Icarus Verilog

• 2D Helmholtz FDFD wave simulation: validated to a residual of 6.6×10⁻¹⁵

• Terminal execution walkthrough: https://youtu.be/kPoNaqUtTb0

• Live interactive console: https://tropos-1.tropos-arch.workers.dev

I’m actively monitoring this thread. The part I’d be especially interested in having someone pressure-test is the thermal-drift assumption behind the multi-level analog restorer (41.3 ps @ 2.5 GHz).

If you work in hardware, photonics, or mathematical modeling and think there’s a weak point in the assumptions, I’d genuinely like to hear it. I’m happy to go into the raw math, waveforms, or the underlying implementation.

Description of subprojects and results, including major changes from the original proposal

While we fell short on our fundraising goal, we have continued to plan and implement new workshops (one in January 2026 and the next in September 2026). The donations we received helped us fund these workshops and we thank everyone who contributed.

Spending breakdown

All funds were used to pay our staff who are working on new workshops.

Such awesome work! So proud of the Japan chapter.

@amiralia Hi, yes this was sent by ACH on November 2.

I haven't read the proposal in detail but Caitlin Yardley is really talented at what she does and a pleasure to work with!

She's especially great at talking to a wide variety of creators so would be a great fit for an agency.

Description of subprojects and results, including major changes from the original proposal

Description of subprojects and results, including major changes from the original proposal

The ACX Grants 2024 award significantly strengthened our capacity for bacteriophage research and development in Nigeria. The funding allowed us to improve our laboratory infrastructure, expand our phage collection, support postgraduate research, and advance several projects from phage isolation toward characterization and potential therapeutic or agricultural applications.

One of the most important infrastructure achievements was the acquisition of an ultracentrifuge. This addressed a major limitation in our laboratory by improving our ability to concentrate and purify bacteriophages locally. Unlike expenditure on individual experiments, this is a long-term investment that will continue to support multiple phage projects, students, and collaborators beyond the grant period.

A major scientific outcome was our work on Salmonella. We isolated and characterized two novel lytic phages, Jerseyvirus ijeoma and Apdecimavirus ayanbimpe, active against Salmonella Typhi. The work progressed to peer-reviewed publication in 2026 in the study “Exploring lytic Salmonella phages as potential alternatives to antibiotics: isolation, characterization, and stability assessment of Jerseyvirus ijeoma and Apdecimavirus ayanbimpe” (Olorundare et al., 2026). The study demonstrated their lytic activity and assessed their stability under different conditions. These phages now provide candidates for the next stage of our work, particularly preclinical evaluation of safety and therapeutic efficacy.

Our work on Pseudomonas aeruginosa also expanded considerably. An MSc research project resulted in the isolation of 34 bacteriophages active against P. aeruginosa, contributing to our growing phage collection. We also completed genomic characterization of the lytic P. aeruginosa phage CSSBELLO(Igwe, Blessing Chidinma, Adesola Olaleka, and Nnaemeka Nnadi. "Isolation and Genomic Characterization of a Novel Lytic Bacteriophage (CSSBELLO) Infecting Multidrug-Resistant Pseudomonas aeruginosa, Isolated from Wastewater in Jos." Frontiera Scientia 1.1 (2026): 10-19, which was subsequently reported in a peer-reviewed publication. These studies have strengthened our ability to move from phage isolation through biological characterization to genome-level analysis. This work is also feeding into our research on a phage-based bandage for infected wounds, particularly as a potential delivery approach for difficult-to-treat wound infections.

Another important development was our expansion into phage-based vaccine research. We developed a multi-serotype Salmonella phage-based vaccine construct intended to provide broader protection across different Salmonella serotypes. Candidate components have undergone computational and experimental work, with in vivo evaluation representing the next major stage of development.

The programme also expanded beyond human health into agricultural phage biotechnology. In our Xanthomonas project, five bacterial isolates were obtained, three of which were demonstrated to be pathogenic. We subsequently isolated seven lytic bacteriophages, with three showing activity against all three pathogenic isolates tested. Plaque assays, enrichment and purification have been undertaken, and the strongest candidates provide the basis for developing a phage cocktail for future evaluation against bacterial disease in tomato. This has broadened the programme toward a One Health approach by applying the same phage expertise to both human health and food production.

We have also strengthened our capacity in phage genomics and bioinformatics. An important outcome has been the development of PhageMine, an open-source bioinformatics workflow for bacteriophage genome analysis. This development helps address one of the challenges we encountered as our phage collection expanded: the need to move beyond isolation and phenotypic characterization to systematic genomic analysis of candidate phages.

The grant has additionally supported a wider capacity-building effect. MSc and PhD students have participated in projects involving Salmonella, Pseudomonas, phage-based wound applications, and agricultural phages. The laboratory infrastructure and expertise developed through these activities are therefore supporting a growing group of researchers rather than a single project.

Major changes from the original proposal

The overall objective of building local capacity for phage research and developing phage-based solutions has remained unchanged. However, the programme became broader than originally anticipated. Our initial emphasis was largely on phage isolation, characterization, storage and therapeutic development. During implementation, this expanded to include phage-based vaccines, wound-delivery applications, agricultural biocontrol, genomic analysis, bioinformatics and postgraduate training.

We also refined our approach to clinical translation. Rather than moving directly from promising laboratory results toward clinical trials, we now recognize the importance of a staged pathway involving comprehensive genomic characterization, candidate selection, preclinical safety and efficacy studies, manufacturing and quality control, regulatory engagement and, ultimately, appropriately approved clinical evaluation.

Overall, the ACX Grants 2024 award has produced both immediate research outputs and longer-term research capacity. The acquisition of the ultracentrifuge, expansion of our phage collection, publication of characterized phages, development of new therapeutic and agricultural applications, bioinformatics capacity and training of postgraduate researchers represent outcomes that will continue beyond the grant period.

Our next phase will build on this foundation by advancing the strongest phage candidates into preclinical studies, further developing the Xanthomonas phage cocktail and phage-based bandage, strengthening the phage bank, and pursuing the manufacturing and regulatory capacity required to translate promising phages into practical applications.

I'm taking this off tomorrow if you guys won't fund it. Have a good time

@raber3 Wooo thanks for your support!

Since the original proposal, the experiments have pushed the question beyond simply asking how much information can be recovered from a transformer's own write directions. Across five models, native writes could often be followed downstream even after their original directions became difficult to recover from the accumulated residual, with their transported descendants retaining information about both their source and later functional effects. This points to a useful separation between where a computation is written, how it is transformed through the residual stream, and what part of it remains functionally relevant.

More broadly, the experiments suggest that residual perturbations can become physically high-dimensional while their behaviorally relevant structure remains comparatively low-dimensional, and that provenance and causal importance are not interchangeable. Despite the grant auto-rejection, I’m releasing these results and the associated negative findings as an extension of the original project in the hope that they’re useful for people studying circuit tracing, activation transport, and the relationship between parameter-level structure and learned representations.

Interesting project--approved!

Since publishing the ILRL Public Evaluation and Findings Project, I have created a public ILRL introduction and organized the method into a simple continuation path:

Conversation → editable review → human-AI interaction receipt → portable checkpoint → later continuation → outcome receipt

Public site: https://innovation-lineage-record-live.rrbberkovich.chatgpt.site/

I have also prepared public documentation for participation, evidence states, adaptive connections, contribution history, and correction handling. These materials are intended to support a small, privacy-safe outside-user evaluation. They do not establish independent validation yet.

The next measurable step is to invite a small number of adult participants to complete one continuation cycle. I will preserve what they received, what they understood, what confused them, what they corrected, and whether the checkpoint helped them continue later. Successful and unsuccessful outcomes will both remain part of the findings.

The Manifund grant would support the work described in the proposal: a sanitized public evaluation corpus, outside-reader sessions, adversarial cases, scoring, accessible demonstrations, privacy review, and a public findings report. Private conversations, credentials, and private lineage archives will not be published.

Current evidence state:

ESTABLISHED: the public site and core workflow exist.

READY: participant-protection and evaluation documents are being prepared.

OPEN: outside-user continuation results and Manifund funding.

NOT_MEASURED: broad usability, independent validation, market demand, and commercial value.

I welcome questions about the evaluation design, privacy boundary, scoring plan, or intended public benefit.

— Richard Ryan Berkovich

[]

BASE is doing an important work in AI Safety and needs as much support as possible.

I think you do a great work, Anna, and will be happy to donate again this year.

I think this is an interesting project -education is something that for personal and professional reasons resonates with me- and I have high trust in Jack, so I'll likely be contributing to it soon.

Tandena, I think this is great work. Thank you for doing it. I'm glad my offer to pay enticed you to do this. Post this on the EA forum?

@MarcusAbramovitch Here's the "vulture carcass disposal x human health" research proposal

Rubi was our mentor during the PRISM fellowship program when we were investigating AI preference drift during training. He gave good direction without micromanaging, which meant we could move fast and weren't dependent on him to make progress. He was also great at guiding us where to focus when needed. I have no doubt he would be great at managing a team at Principled Agents and hope to work with him again in the future.

Description of subprojects and results, including major changes from the original proposal

The project was completed and final outputs submitted to UK ARIA. The research questions I studied here formed the inspiration and basis for my continued work during the PIBBSS fellowship in 2026.

@aCFAR Not super related but wanted to say, we rented out CFAR's Bodega Bay Venue for a midpoint retreat for the Surplus incubator. It was fantastic; I'd highly encourage other folks considering doing eg a weekend retreat or workshop, to look into renting their space: https://www.rationality.org/venue

More related: many of the founders expressed an interest in participating in a future CFAR workshop. Keep up the good work!

FloraForge’s new website is now live, with an overview of the platform, current validation work, and long-term vision:

https://www.floraforgelabs.com/

Hi all — Sina here, lead architect on TROPOS-1.

A quick upfront note on the 87% Pangram flag: as an independent researcher whose primary language isn't English, I used LLMs to help polish and structure the written prose of this grant proposal.

The underlying architecture, RTL, mathematical formulation, and validation work are my own. For anyone who wants to verify rather than take that on faith:

• 87-page master dossier + executed logs: archived on Zenodo (DOI: 10.5281/zenodo.22637552)

• Synthesizable SystemVerilog RTL: passing self-checking testbenches under Icarus Verilog

• 2D Helmholtz FDFD wave simulation: validated to a residual of 6.6×10⁻¹⁵

• Terminal execution walkthrough: https://youtu.be/kPoNaqUtTb0

• Live interactive console: https://tropos-1.tropos-arch.workers.dev

I’m actively monitoring this thread. The part I’d be especially interested in having someone pressure-test is the thermal-drift assumption behind the multi-level analog restorer (41.3 ps @ 2.5 GHz).

If you work in hardware, photonics, or mathematical modeling and think there’s a weak point in the assumptions, I’d genuinely like to hear it. I’m happy to go into the raw math, waveforms, or the underlying implementation.

Description of subprojects and results, including major changes from the original proposal

While we fell short on our fundraising goal, we have continued to plan and implement new workshops (one in January 2026 and the next in September 2026). The donations we received helped us fund these workshops and we thank everyone who contributed.

Spending breakdown

All funds were used to pay our staff who are working on new workshops.

Such awesome work! So proud of the Japan chapter.

@amiralia Hi, yes this was sent by ACH on November 2.