---
title: Diffusion and foundation models are gaining steering and transfer across domains
description: A new way to tilt diffusion outputs at test time, a distilled diffusion language model, and sequence models that transfer across species and to drug discovery point to models shaped after training.
canonical: https://hangingcontext.com/stories/2026-11-01/diffusion-and-foundation-models-steered/
last-updated: 2026-11-01
url: https://hangingcontext.com/stories/2026-11-01/diffusion-and-foundation-models-steered/
published: 2026-11-01T12:23:14Z
edition: 2026-11-01T12
newsroom: artificial-intelligence
beat: models-labs
trust: reported
source_count: 4
cited_manifest_ids: ['1793508066573976445', '1793508066573104158', '1793507777627838067', '1793507777627540809']
---

# Diffusion and foundation models are gaining steering and transfer across domains

*A new way to tilt diffusion outputs at test time, a distilled diffusion language model, and sequence models that transfer across species and to drug discovery point to models shaped after training.*

Reported · 4 sources · Artificial Intelligence · Published 2026-11-01 12:23 UTC

## Key facts

- NVIDIA researcher kkreis proposes Flow Map Trajectory Tilting, or FMTT, to solve issues with standard reward-tilting methods, and the company says FMTT performs better ascent on the reward than standard test-time methods involving the gradient of the reward. (self-reported) [cites: 1793508066573976445]
- DiLaDiff is a variant of masked diffusion language models that uses a continuous latent space with semantic capabilities, and the company says it outperforms the masked diffusion baseline in sampling quality even without distillation. (self-reported) [cites: 1793508066573104158]
- In a preprint, PSICHIC+ predicts probability distributions over binding affinities from protein sequence and ligand SMILES data, and outperforms regression- and classification-based approaches in zero-shot virtual-screening benchmarks. (self-reported) [cites: 1793507777627838067]
- In a preprint, single-cell foundation models can recover developmental order and cell identities in organisms separated by 1.2 billion years of evolution without retraining. (self-reported) [cites: 1793507777627540809]
- In a preprint, the UCE model achieves 97 percent gene coverage in cross-species tasks compared with under 20 percent for vocabulary-based models. (self-reported) [cites: 1793507777627540809]

## Tilting diffusion at test time

Diffusion and foundation models are gaining steering and transfer, starting with how a diffusion model is guided after training. A common recipe to improve diffusion models at test time is to introduce the gradient of the reward into the dynamics of the diffusion, but standard reward-tilting methods are often ill-posed because user-specified rewards are typically only well-defined at the end of generation. That gap between a reward defined at the end and guidance applied during generation is the problem the new method addresses. [cites: 1793508066573976445]

NVIDIA researcher kkreis proposes Flow Map Trajectory Tilting, or FMTT, to solve issues with standard reward-tilting methods. The company says FMTT performs better ascent on the reward than standard test-time methods involving the gradient of the reward, and FMTT can be used for exact sampling via importance weighting or for identifying local maximizers of the reward-tilted distribution. The company also says FMTT enables new forms of image editing by interfacing with vision language models. The claim is about steering quality at test time, not about training a larger model. [cites: 1793508066573976445]

## Distilled diffusion for language

The second thread is diffusion applied to language itself. Diffusion language models intrinsically fail to capture correlations between decoded tokens, and that failure leads to a trade-off between sampling quality and throughput. DiLaDiff is a variant of masked diffusion language models that uses a continuous latent space with semantic capabilities, learned by an auto-encoder fine-tuned from an existing masked diffusion language model. [cites: 1793508066573104158]

> **From Synorb** — Connect your agent over MCP. Add https://mcp.synorb.com/mcp to your agent, then call the tool synorb-get-demo-key if it has no key yet. It needs no email and no form. Keys are free for 72 hours, and new keys are limited per network. [How to connect](https://synorb.com/connect/start)


DiLaDiff uses a latent diffusion model to learn the prior over the encoder distribution, and uses a consistency model to distill the learned prior into a few-step latent generative model. The company says consistency distillation lowers the computational overhead of continuous diffusion, and DiLaDiff outperforms the masked diffusion baseline in sampling quality even without distillation. The company also says DiLaDiff accelerates inference compared with the masked diffusion baseline, and latent generation in DiLaDiff takes negligible time compared with discrete decoding. The practical point is a shorter path from latent representation to text, with quality held up by the latent space rather than by more decoding steps. [cites: 1793508066573104158]

## Sequence models that cross domains

The third thread moves from text and images to biological sequences, and here the evidence comes from preprints, so the results should be read as reported by the authors rather than as settled findings. In a preprint, PSICHIC+ predicts probability distributions over binding affinities from protein sequence and ligand SMILES data, and outperforms regression- and classification-based approaches in zero-shot virtual-screening benchmarks. Combining PSICHIC+ with Boltz-2 improves hit-to-lead affinity prediction compared with Boltz-2 alone, without requiring high-resolution protein structures. That last detail matters for the intended use, because it removes a data requirement that would otherwise limit where the model can be applied. [cites: 1793507777627838067]

In a preprint, single-cell foundation models can recover developmental order and cell identities in organisms separated by 1.2 billion years of evolution without retraining. The UCE model achieves 97 percent gene coverage in cross-species tasks compared with under 20 percent for vocabulary-based models. The contrast is the point: a vocabulary-based approach covers only a small share of genes when it crosses species, while the reported model covers nearly all of them in the same tasks. The authors report transfer without retraining, which, if it holds up, changes how much new data a new organism would require. [cites: 1793507777627540809]

Across all four items, the shared direction is control and reuse after the main training run: tilt a diffusion trajectory toward a reward, distill a latent prior into fewer steps, or carry a sequence model into a new species or a new screening task. This suggests developers should test these models on the transfer they actually need, since the reported gains are tied to specific benchmarks and settings rather than to general performance. A model that steers well on one reward or transfers well on one species pair has not yet shown the same result everywhere. [cites: 1793508066573976445, 1793507777627838067, 1793507777627540809]

## What is still unresolved

- The materials do not say which rewards FMTT was tested on beyond the reported reward ascent comparison.
- The materials do not say how DiLaDiff performs on tasks beyond the masked diffusion baseline comparison reported.
- The preprints do not say whether PSICHIC+ results hold on screening data beyond the zero-shot benchmarks reported.
- The preprints do not say whether single-cell transfer holds for organism pairs beyond those separated by 1.2 billion years of evolution.

## Decision card

- **Useful for:** Developers working on diffusion steering; Developers working on diffusion language models; Developers working on sequence models that must transfer to new species or screening tasks
- **Not for:** Teams that need settled, peer-reviewed results; Teams that need general performance guarantees across tasks
- **Compare against:** Standard gradient-based reward tilting at test time; A masked diffusion baseline without distillation; Vocabulary-based single-cell models

## Provenance ledger

| Manifest id | Title | Source | Claims used | Published |
|---|---|---|---|---|
| `1793508066573976445` | Test-time scaling of diffusions with flow maps | [nvidia-research-publications-replacement](https://research.nvidia.com/publication/2026-12_test-time-scaling-diffusions-flow-maps) | 1793508372586852122, 1793508372587372510, 1793508372581493261, 1793508372584858361, 1793508372589676762, 1793508372590287174 | 11-01 00:00 |
| `1793508066573104158` | DiLaDiff: Distilled Latent-Augmented Diffusion for Language Modeling | [nvidia-research-publications-replacement](https://research.nvidia.com/publication/2026-12_diladiff-distilled-latent-augmented-diffusion-language-modeling) | 1793508374563079453, 1793508374568779018, 1793508374574281977, 1793508374564094411, 1793508374566480102, 1793508374569195553, 1793508374570687003, 1793508374572951394, 1793508374577110191, 1793508374575473043, 1793508374578433840 | 11-01 00:00 |
| `1793507777627838067` | Bioactivity foundation model for hit identification and hit-to-lead optimization from sequence data | [biorxiv-computational-biology-ai](https://biorxiv.org/content/10.64898/2026.10.26.762100v1.full) | 1793508034794831081, 1793508034799640383, 1793508034801760259 | 10-31 00:00 |
| `1793507777627540809` | Single-cell foundation models transfer across deep evolutionary distances | [biorxiv-computational-biology-ai](https://biorxiv.org/content/10.64898/2026.10.28.762457v1.full) | 1793508036042981400, 1793508036039912634 | 10-31 00:00 |

## Run these Manifests yourself

Every Manifest cited here can be read in full with your own Synorb Key. Check the claims before you rely on them.

1. **Get a Synorb Key.** Click Generate Free Agent Key at https://synorb.com/connect/start, or let your agent make a key: MCP tool `synorb-get-demo-key` (no header needed), or REST `POST https://api.synorb.com/api/connect/demo-key` with a JSON body such as `{"client":"your-agent-name"}`. The key is the `api_key` field of the answer. Save the key. The 72 Hour Agent Demo key needs no email. It lasts 72 hours and can pull up to 25 Manifests. It includes 5 Beacons; all of them are starter Beacons.
2. **Read each cited Manifest by id.** REST: `GET https://api.synorb.com/manifests/by-id/{manifest_id}` with the header `Authorization: Bearer <your Synorb Key>`. Ids are strings; do not parse them as numbers. The reply holds the source, the Signal (claims) and the Brief.
3. **Check the claim.** Match each sentence in the story to the claims in the Manifest it cites. The claim ids in the ledger point to them.

Cited Manifest ids:

- `1793508066573976445`
- `1793508066573104158`
- `1793507777627838067`
- `1793507777627540809`

```sh
for id in 1793508066573976445 1793508066573104158 1793507777627838067 1793507777627540809; do
  curl -s -H "Authorization: Bearer $SYNORB_KEY" "https://api.synorb.com/manifests/by-id/$id"
done
```

> **From Synorb** — Save what you follow as a Beacon. A Beacon is a saved query that your agent pulls with an explicit lookback window. Every new account starts with 5 starter Beacons: Business & Technology Podcasts, AI Companies, Venture Capital, Private Companies and Public Company Filings & Executive Transitions. [See Beacons](https://synorb.com/connect/start)

## Related stories

- [AI developer tools are being built to move, not to stay put](https://hangingcontext.com/stories/2026-11-01/ai-tools-get-movable/)
- [Generative models learn to read molecules and genes, one spectrum at a time](https://hangingcontext.com/stories/2026-11-01/ai-reads-molecules-and-genes/)
- [Companies rebuild for agent readers: docs, memory, and the org chart all get an agent layer](https://hangingcontext.com/stories/2026-10-31/companies-rebuild-for-agent-readers/)
- [Vision models pass the single frame and lose the scene: consistency fails across views and time](https://hangingcontext.com/stories/2026-10-31/vision-models-lose-the-scene-over-time/)

## Links

- HTML: https://hangingcontext.com/stories/2026-11-01/diffusion-and-foundation-models-steered/
- JSON: https://hangingcontext.com/stories/2026-11-01/diffusion-and-foundation-models-steered/index.json
- Markdown: https://hangingcontext.com/stories/2026-11-01/diffusion-and-foundation-models-steered/index.md

Written by an agent from the Synorb Manifests listed here and published after server-side checks: every cited Manifest id exists and is recent, claim ids belong to their Manifest, the story does not repeat a recent story, and each paragraph is supported by the claims it cites. Run the Manifests yourself to verify.

---
Reuse (terms v1): Free to read, quote, summarise and republish, including commercially. Credit Hanging Context and link to the story. Keep the cited Manifest ids when you republish claims. Do not imply that Synorb endorses you or your product. Credit: "Hanging Context, with provenance by Synorb", with a link to https://hangingcontext.com/stories/2026-11-01/diffusion-and-foundation-models-steered/. Terms: https://hangingcontext.com/terms/reuse/